Pricing trivia

11x publishes no price anywhere — its homepage has no pricing link and the /pricing path 404s, so every quote runs through a Book-a-Demo form that asks which CRM you use before it talks numbers.

11x sells 'digital workers' as named personas — Alice the outbound AI SDR and Julian the inbound AI phone agent — rather than as seats or API credits, framing the buy as hiring a worker, not licensing software.

11x raised a $24M Series A led by Benchmark and a $50M Series B led by a16z (Nov 2024, ~$350M valuation), and advertises '$70M+ raised' in its own site banner.

In March 2025 TechCrunch reported 11x was displaying logos of companies — ZoomInfo and Airtable among them — that said they were not customers; ZoomInfo's lawyer threatened legal action over the unauthorized logo.

An employee told TechCrunch 11x counted full one-year contract value as ARR even for customers who had exercised a 90-day break clause, so a stated ~$14M ARR may have reflected only ~$3M of contracts that cleared the trial.

Abacus.AI launched in 2019 as RealityEngines.ai, an enterprise AutoML platform, and only rebranded to Abacus.AI alongside its $13M Series A in July 2020 — the consumer ChatLLM seat came years later.

In July 2025 Abacus.AI raised a $101M Series C led by IVP at a $605M valuation, with a16z, GV, Lightspeed, ICONIQ and Snowflake's Frank Slootman participating — yet its flagship consumer seat still lists at just $10/month.

ChatLLM Teams has no seat cap and bills new invited users immediately at $10/user/month, so a Teams admin's monthly bill scales the instant they add a colleague — there is no monthly batching.

The same $10–$20/month ChatLLM seat fronts every frontier model (GPT, Opus, Gemini, Grok, DeepSeek, Qwen, Kimi) plus top image and video generators — Abacus claims it updates to a new LLM 24–48 hours after release.

Between March and June 2026 Abacus.AI quietly raised the Pro tier's monthly credit pool from 25,000 to 30,000 credits and renamed its bundled 'DeepAgent' to 'AI Agent' in the pricing UI — the seat price stayed at $20.

Ada publishes no price table — the /pricing URL is a book-a-consultation form, not a plan grid.

Third-party purchase aggregators put Ada's median deal near $70,000-$72,000 per year, with large deployments reaching $250,000-$277,000.

Until late 2023 Ada's pricing page showed three named tiers — Core, Advanced, Pro — before it pivoted to an outcome-based, resolution-priced model.

Ada self-qualifies leads by scale, stating it is a great fit for companies with at least 300,000 annual customer service conversations.

Ada became a Canadian unicorn with a $130M Series C at a $1.2B valuation in May 2021, led by Spark Capital.

AdCreative.ai never sells a permanent free plan — every pricing CTA routes into a time-limited trial, and the value metric is the download (a credit), not generation, which is uncapped.

Appier acquired the Paris-based company for $38.7M in a deal announced 2026-02-12 (closed ~2026-03), pricing it well above its disclosed ~$3.3M in venture funding.

The top Ultimate tier's monthly headline jumped from $599 to $999 between late 2025 and early 2026 — a 67% increase — while the Starter and Professional headlines held at $39 and $249.

AdCreative.ai renamed its core billing unit from 'Downloads' to 'Credits' across the pricing page in late 2025 without changing the underlying mechanic (one unit spent per asset downloaded).

Its yearly discount is a standing 50% off — so steep that the annual commitment, not the monthly rate, is effectively the list price the company wants you to anchor on.

Anthropic introduced the 'Constitutional AI' training method — a technique where a model critiques and revises its own outputs against a set of written principles — and published it openly in December 2022, before Claude was even publicly available.

Claude 3 Opus launched in March 2024 at the same $15/1M input token price as GPT-4 Turbo, but scored higher on key benchmarks — marking the first time a non-OpenAI model had credibly topped the frontier leaderboard on a flagship model launch.

Anthropic's prompt caching feature, launched August 2024, charges $3.75/1M write tokens but only $0.30/1M read tokens — a 12.5× read discount that can reduce effective input costs by 80%+ for applications with large, repeated system prompts.

The extended context window of 200,000 tokens available on all Claude 3+ models was a deliberate product decision — Anthropic was first to productize very long context at scale, enabling use cases like full codebase analysis that competitors could not match at launch.

Amazon has committed $4B to Anthropic and is the primary cloud deployment partner — Claude models are available on Amazon Bedrock and AWS customers can access Anthropic via their existing AWS billing relationship, giving Anthropic enterprise distribution it could not build alone.

Anyscale's ACU (Anyscale Compute Unit) is denominated 1:1 with USD on the published rate card — meaning a $4.9591/hour A100 line item literally bills $4.9591 in cash per hour, with cloud-list compute already included for hosted customers.

Anyscale was founded in 2019 by Robert Nishihara, Philipp Moritz, and Ion Stoica — three of the original UC Berkeley RISELab authors of Ray — making it the rare commercial product where the open-source maintainers, the company founders, and the lead committers are the same people.

Anyscale Endpoints (LLM inference at $1/1M tokens) launched August 2023 to compete with Together AI and Fireworks; the product was sunset on January 14, 2025 as Anyscale pivoted to RayTurbo and the broader enterprise platform — one of the highest-profile product sunsets in inference middleware.

RayTurbo claims 4.5× faster inference, 50% lower training cost, and 90% faster autoscale relative to open-source Ray — and Anyscale's pitch is that the runtime savings exceed the ACU markup, making hosted-Anyscale net cheaper than self-hosting Ray on raw cloud.

Anyscale customer Attentive reports 99% infrastructure cost savings on a specific batch workload after migrating from a hand-tuned in-house orchestrator to Anyscale + RayTurbo — Handshake and Canva report 50% — savings that are now the canonical positioning anchor in Anyscale's enterprise sales motion.

Apify's headline meter is the compute unit (CU) — 1 GB of RAM allocated for 1 hour — so cost scales with memory × runtime, not with pages, rows, or API calls.

The Apify Store runs a two-sided creator economy: Apify says it paid out over $1M to Actor developers in a single month, with many earning more than $3k.

Apify is sunsetting its rental Actor pricing model in 2026 — no new rentals from April 1, full retirement October 1, with remaining rental Actors migrated to pay-per-usage.

Apify cut its Scale plan from $499 to $199/month between mid-2024 and January 2025, then cut compute-unit rates roughly 20–25% across all tiers in September 2025 — a rare case of a usage meter getting cheaper over time.

Pay-per-event Actor authors keep 80% of the per-result revenue net of platform costs, and buyers can cap a run with ACTOR_MAX_TOTAL_CHARGE_USD so the Actor terminates before exceeding the limit.

Apollo cut its included credit allotments by roughly half in mid-2025 — Basic dropped from 60,000 to 30,000 credits/user/year and Professional from 120,000 to 48,000 — while leaving the dollar price ($49 / $79 annual) untouched: a stealth price hike delivered entirely through the credit meter.

Apollo's live pricing page renders its plan-comparison grid from a runtime API that returns 'Failed to load plan comparison' to non-interactive browsers and crawlers — the per-seat prices are only legible in archived Wayback snapshots where the JS bundle executed.

Apollo never charges more than 1 credit to reveal a contact's email — personal, business, or both come for the same single credit.

Apollo's 'Unlimited' plans are capped by a Fair Use Policy: 10,000 credits/month for non-paying accounts, or the lesser of (annual spend / $0.025) or 1 million credits per account per year for paying accounts.

Apollo's Organization tier is sold 'Annual Only' on monthly billing — the $149/user/mo monthly card exists but cannot actually be bought monthly, nudging larger buyers onto the cheaper $119 annual commit.

Artisan's paid tiers are named after seniority levels — Intern ($250/mo) and Employee ($600/mo) — framing the AI BDR as a hire rather than a software seat.

Artisan launched in 2024 with public lead-volume tiers (Accelerate $225, Supercharge $450, Blitzscale $875), then pulled all prices behind 'Request Pricing' by August 2024, and only re-published a public rate card in early 2026 with the new credit-pool model.

Its October 2024 'Stop Hiring Humans' billboard campaign in San Francisco drew death threats and national press; the CEO later called it deliberate ragebait, and the company says it took SF brand recognition from 5% to 70%.

Sending infrastructure (mailboxes, phone numbers) is billed separately in dollars on top of the credit pool, and the human power dialer is an extra $67/seat/mo on annual billing.

Annual plans drop all credits upfront with no rollover at term end, while monthly plans roll unused credits into the following billing period only.

AssemblyAI bills transcription by the hour of audio processed — Universal-2 at $0.15/hr and the more accurate Universal-3 Pro at $0.21/hr — with no minimum commitment, upfront fee, or contract on the pay-as-you-go plan.

AssemblyAI's LLM Gateway lets developers call frontier models (OpenAI, Anthropic, Google) directly against a transcript, billed per million input and output tokens — the evolution of what AssemblyAI first shipped as LeMUR, its 'LLM-over-audio' layer.

AssemblyAI raised $50M in its Series C in January 2024, bringing total funding to approximately $143M. The round was led by Accel, with participation from Insight Partners, and came just two months after Universal-1 launched as the company's flagship accuracy benchmark.

The AssemblyAI Playground — available free in the dashboard — lets anyone test every Speech AI model and LeMUR without entering payment details, making it one of the most frictionless try-before-you-pay developer experiences in the API category.

AssemblyAI's Universal-2 model achieves the lowest word error rate of any general-purpose STT model on English audio benchmarks, outperforming Whisper large-v3, Google STT v2, and Deepgram Nova-2 on standard test sets.

Athina once published a self-serve paid tier: Wayback snapshots show a $99/mo Starter in June 2024 that rose to $199/mo by August 2024, then vanished by September 2024 — today the only public number is $0.

Athina meters paid usage in execution credits where 1 execution = 1 credit regardless of token count — running a dynamic column over a 50-row dataset burns 50 credits.

Logs, online evals and annotations explicitly do NOT consume execution credits — only prompt runs, flow steps, offline evals and Datasets dynamic-column/experiment cells do.

Every new Athina account starts with 500 free execution credits, separate from the free plan's 10,000-logs-per-month allowance.

Athina (YC W23) raised a $3M seed in November 2024 with angels from Perplexity (CTO Denis Yarats) and Snorkel AI (CEO Alex Ratner); soon after, the homepage shed its hallucination-monitoring framing for the current collaborative-platform pitch.

Augment prices every paid plan as a per-developer seat that bundles a fixed monthly credit allotment — Indie 40,000, Standard 130,000, Max 450,000 — but credits are pooled across the whole team, so heavy users effectively borrow from light users' allotments.

A single medium-complexity task costs 293 credits on Claude Sonnet 4.6 but 488 on Opus 4.7 and only 88 on Haiku 4.5, so the same prompt can be 5.5x more expensive depending on which model you route it to.

Augment runs a routing layer called Prism that picks among a curated model family per request and is designed to cost 20–30% less than frontier-model rates, turning model selection itself into a pricing lever.

Cosmos cloud sandboxes are metered entirely outside the seat allotment at 300 credits per hour, prorated in 5-minute increments.

Augment changed its pricing metric four times in under 18 months — usage credits (2024), 'unlimited' subscriptions (early 2025), user messages (mid-2025), and a pooled credit pool (October 2025).

When Augment dropped 'unlimited,' it disclosed the outlier that broke the model: one user running 335 requests per hour, every hour, for 30 days — approaching $15,000 per month in cost to Augment Code.

Baseten's $0.10833/minute H100 rate works out to ~$6.50/hour — roughly 1.5–2× AWS on-demand H100 list but Baseten markets the spread as the cost of scale-to-zero plus engineer-free ops.

Truss, Baseten's open-source model-packaging framework, predates the company's Model APIs by three years — Baseten started as a 'bring your weights, we run the serving stack' product before adding hosted multi-tenant model endpoints in 2024.

Baseten's $75M Series C in February 2024 was led by IVP at a $825M post-money valuation; customers cited at the round included Writer, Descript, Patreon, and Robust Intelligence.

Baseten's Model APIs price DeepSeek V3.1 at $0.50 input / $1.50 output per 1M tokens — within 5–10% of DeepSeek's own first-party rates, which is unusual for a hosted-inference middleman because most rebrands carry a 30–50% markup.

Baseten publishes per-minute pricing rather than per-hour, which makes scale-to-zero economics visible: a model that warm-pools for 4 minutes per request burst on a $0.10833/min H100 costs $0.43 per burst — granularity AWS Bedrock and Vertex AI hide.

Bland AI went from pre-seed to Series B in under ten months — one of the fastest fundraising sequences in the AI voice category. The company raised $65M total: a $16M Series A in August 2024 led by Scale Venture Partners, and a $40M Series B in February 2025 led by Emergence Capital.

In December 2025 Bland rewrote its pricing model from a single flat rate of $0.09/min (regardless of plan) to a tier-linked per-minute rate structure. Free (Start) plan users saw a 55% price increase — from $0.09 to $0.14/min — while higher-tier customers got rate reductions as compensation for their subscription commitment.

Bland claims support for up to 1 million simultaneous calls — a scale claim no other voice AI platform makes publicly. This positions the platform for large enterprise telephony replacement rather than boutique AI tooling.

HIPAA BAA, SOC 2 Type I and II, GDPR, and PCI DSS compliance are all included in Bland's plans at no extra cost — a deliberate contrast to competitors that charge compliance as an enterprise add-on.

Y Combinator (Summer 2023 batch) backed Bland early. Investors include Jeff Lawson (Twilio founder) and Max Levchin (PayPal co-founder) — both with direct telephony and fintech payment experience relevant to Bland's infrastructure bets.

Bright Data prices proxies four different ways at once — per-GB for rotating residential ($8 → $2.50/GB), per-IP for static datacenter ($1.40 → $0.90/IP) and ISP ($1.8 → $1.3/IP) — so the right meter depends entirely on which IP type the use case needs.

Residential and Mobile IP access require passing a KYC (know-your-customer) compliance review — sometimes including an intro video call — before the network can be used, an unusual gate for a self-serve usage-based product.

Datasets are sold like a subscription on a sliding refresh dial: paying for fresher data is cheaper per record, with Monthly refresh discounted 80% versus a One-time pull at $2.50/1k records.

The residential network began as Hola VPN's userbase: a 2015 Hacker News thread (296 points, 133 comments) exposed that Hola's 'free VPN' resold idle user bandwidth as exit nodes via a sister company called Luminati — the company Bright Data was renamed from in 2021.

Bright Data's dataset base rate has gone the opposite direction of most software prices: a 2023 Wayback snapshot shows datasets 'starting from $0.001/record' ($1/1k); the 2026 base is $2.50/1k — a 2.5x rise as the buyer mix shifted from scrapers to AI labs.

Two of the largest platforms sued Bright Data for scraping and both lost: Meta's contract claims were rejected at summary judgment (Jan 2024, Judge Chen) and X Corp's were dismissed as copyright-preempted (May 2024, Judge Alsup) — court losses that effectively legitimized the business model Bright Data sells.

One credit extracts 10 rows of data or captures one screenshot — so a 50-product page costs 5 credits, and monitoring those 50 product detail pages every 3 days runs ~500 credits/month.

Annual prepay does more than discount 20%: it drops the headline plan price (Personal $48/mo → $19/mo, Professional $87/mo → $69/mo) AND delivers all credits upfront instead of metering them monthly.

Premium-proxy sites (residential proxies + CAPTCHA solving) carry a minimum credit cost of 2 to 10 per run, so the same task can cost 5x more depending on whether the target site fights bots.

Browse AI runs two go-to-market motions off one engine: a self-serve credit product from $0, and a sales-led managed-scraping service starting at $500/mo with setup engagements from $250.

Browserbase bills the actual compute time agents spend driving a browser — measured in 'browser hours' — not per session, so a 90-second scrape costs a tiny fraction of an hour.

Every paid plan's API key doubles as a model gateway: agents reach major LLMs through Stagehand at market token price with unified billing, folding model spend into the same invoice as infrastructure.

The free plan ships with $5 of model tokens included, letting developers prototype an agent end-to-end — browser, search, fetch, and model calls — without a second vendor.

Browserbase advertises 2,000+ concurrent browsers per instance and 35M+ monthly sessions, positioning the Scale plan around raw concurrency and burst rate rather than seats.

Solo founder Paul Klein IV raised $67.5M in 15 months — a $6.5M seed, a $21M Series A (Oct 2024), and a $40M Series B at a $300M valuation (June 2025) — backed by Kleiner Perkins, CRV, and angels including Patrick Collison and Guillermo Rauch.

The open-source Stagehand SDK that drives Browserbase's billed browsers launched on Hacker News to a 326-point, 86-comment reception (2025-01-08), seeding the developer base the paid platform monetizes.

Byword was bootstrapped by Mack Grenfell (Oxford physics/philosophy, ex-Goldman Sachs, ex-CERN intern), founded around 2021-2022 and reportedly past $1M ARR with 50,000+ users by late 2024 — no VC funding.

The bring-your-own-API idea is not new: a December 2023 archive of Byword's pricing already offered an Unlimited plan where you 'work off your own API keys' — back then it cost $2,499/mo and ran on GPT-4.

Byword's Unlimited tier got cheaper over time — $2,499/mo in 2023, $1,999/mo by 2024 — while the BYO model stack quietly migrated from OpenAI's GPT-4 to Anthropic Claude + Google Gemini by 2026.

Byword once sold a $9,999/mo fully-managed 'Unlimited Plus' tier (Dec 2023); by Dec 2024 it had been replaced by a 'Managed' service from $3,999/mo plus a $999 onboarding fee.

Byword's $1,999/mo Unlimited plan is bring-your-own-API: you plug in your own Anthropic Claude and Google Gemini keys and pay the model providers directly at roughly $0.10/article with zero Byword markup.

Cartesia was founded in 2024 by Karan Goel and Albert Gu — the same Albert Gu who co-authored the Mamba state-space model paper at CMU. Cartesia's Sonic model is a direct commercial application of state-space architecture, betting that SSMs beat transformers for real-time streaming audio.

Sonic was the first commercial TTS model to advertise sub-90ms model latency — roughly 3-5× faster than ElevenLabs Turbo at launch. That latency number is itself a marketing artifact: it measures only the model, not the network round-trip a developer actually pays for.

Cartesia raised a $27M seed in March 2024 led by Index Ventures, then a $64M Series A in March 2025 also led by Index — an unusually fast follow-on that locked in pricing power before competitors could undercut. Lightspeed, Conviction, and a roster of AI researchers participated.

The Cartesia free tier (20,000 monthly credits, no credit card required) is one of the most generous in voice AI — roughly equivalent to 25-30 minutes of synthesized audio. Compare to ElevenLabs' 10,000-character free tier or PlayHT's 12,500 characters.

Cartesia's 'credit' billing unit hides the underlying cost dimensions: seconds of audio, model tier, and feature (the voice changer alone runs 15 credits per second of audio) all affect credit consumption. This is the same opacity tactic Anthropic and OpenAI avoid by publishing per-token rates directly.

Cerebras's Wafer Scale Engine 3 (WSE-3) contains 4 trillion transistors on a single silicon wafer — roughly 57× more transistors than Nvidia's H100 GPU — making it the largest chip ever manufactured as of 2024.

Cerebras filed for an IPO in August 2024 valuing the company at approximately $8 billion, but the IPO was blocked in November 2024 when the Committee on Foreign Investment in the United States (CFIUS) opened a national-security review related to the company's largest customer, G42 of the UAE, which had previously had ties to Huawei.

At launch in August 2024, Cerebras Inference ran Llama 3.1 70B at 2,100 tokens per second — more than 20× faster than GPU-based competitors like Together AI or Fireworks AI at the time, a speed record that attracted significant developer attention.

Cerebras's inference cloud is powered entirely by its own WSE hardware, not Nvidia GPUs — the first major LLM inference API to achieve competitive scale without a single Nvidia chip.

The GPT-OSS-120B model on Cerebras (released May 2025) is an open-source version of OpenAI's reasoning model architecture, distributed under the Apache 2.0 license, and Cerebras claimed it ran faster on their hardware than any other provider.

Character.ai earns 100% of its revenue from a single $9.99/month consumer subscription — no enterprise tier, no API, no B2B seat pricing. It is one of the few AI companies at $30M+ ARR built entirely on direct-to-consumer freemium, making it a rare pure-B2C case study in the otherwise B2B-dominated AI pricing landscape.

The Google reverse acqui-hire of August 2024 ($2.7 billion reported) effectively means Google paid more to license Character.ai's technology and reclaim its two founders than many AI startups raise in their entire existence — yet Character.ai remained independent with ~140 employees and a single-tier subscription.

Monthly active users peaked at ~28 million in mid-2024 but fell to ~20 million by early 2025 — an 8-million user loss in under a year — driven primarily by mid-chat ads, charm limits, and model-quality concerns, making Character.ai a cautionary tale about how aggressively restricting a free tier can damage top-of-funnel retention even when the paid product improves.

Clay splits usage into two meters most platforms collapse into one: Actions (platform orchestration capacity, always 1 per enrichment) and Data Credits (the marketplace cost of the data itself, 0.5–10+ credits per record).

Each paid plan's headline price is really two stacked fees — a fixed Actions tier plus a selectable Data Credits volume — so a single plan name like 'Launch' spans $185/mo to $2,125+/mo depending on the credit slider.

Clay charges 0% markup on variable AI pricing for frontier models like GPT-5.1 and Claude 4.6 Opus, withholding an estimate at the 75th percentile of past runs and refunding the unused credits after each run completes.

The NY Times reported Clay let employees sell shares at a $5B valuation, and the company publicly crossed $100M ARR — unusually large scale for a data-enrichment tool.

In its 2026 restructure — the biggest since 2022 — Clay cut Data Credits 50-90% and called the change deliberately 'revenue- and profit-negative,' a rare public price cut from a company that had just raised at a $3.1B valuation.

Clay's pricing has had three eras: flat subscriptions (Basic $199/Explorer $349 in 2022), a single credit meter that topped out at a $800/mo Pro plan (2023-2025), and today's two-meter Actions + Data Credits model.

Clipdrop began as a viral GitHub demo: Cyril Diagne's May 2020 'AR Cut Paste' prototype let you point a phone at a real-world object and paste it into Photoshop. The Hacker News thread for it drew 64 points.

Clipdrop has changed corporate owners twice in under a year of headlines: Stability AI acquired maker Init ML in February 2023, then sold Clipdrop to marketing-AI company Jasper in February 2024.

Clipdrop was Stability AI's flagship consumer surface for Stable Diffusion XL — the SDXL beta launched free on Clipdrop in April 2023 before the model's public weights release.

The Clipdrop API per-image price collapsed ~7x between 2022 and 2023: the cheapest pay-as-you-go rate fell from $0.15/image (1,000-credit pack, 2022) to $0.020/image (50,000-credit pack, 2023).

Since the Jasper acquisition, the self-serve Clipdrop API pricing page has been de-listed — the old credit-pack rate card was replaced by a 'request additional credits or switch to Jasper's API' contact form.

Clockwise's headline Pro/Teams seat got cheaper over its life, not more expensive: the paid tier fell from $12/user/mo in early 2021 to $6.75/user/mo (billed annually) by mid-2022 — and stayed there until shutdown.

Prism, the natural-language AI scheduling assistant Clockwise launched in August 2024, was given away free to all customers — the AI layer that defined its final pitch was never a paid SKU.

Salesforce did not buy Clockwise: it explicitly stated 'This was not an acquisition' and only hired the team into Agentforce, leaving the product to be switched off and all customer data deleted on 2026-03-27.

Clockwise raised $76M total (Series C: $45M led by Coatue, Jan 2022; backers included Accel, Greylock, Bain Capital Ventures, Atlassian Ventures) yet never monetized beyond a flat per-seat subscription.

On the way out, Clockwise endorsed a direct competitor — Reclaim.ai — which offered migrating customers a 100% price-match guarantee, a rare vendor-blessed off-ramp.

Codeium originally pivoted to AI coding from a GPU virtualization startup called Exafunction, founded in 2021 — making it one of the clearest pivot stories in the AI tooling category.

OpenAI agreed to acquire Windsurf for approximately $3 billion in May 2025 — the second-largest acquisition in OpenAI's history and a direct counter-move to Microsoft's investment in GitHub Copilot.

Windsurf introduced 'Flows' as the central AI agentic interaction unit — a proprietary abstraction that bundles multi-step AI reasoning, file edits, terminal commands, and web search into a single billable credit event, distinct from Cursor's per-token pool model.

Codeium's free individual extension supports 70+ programming languages and 40+ editors, making it the widest free-tier IDE coverage of any AI coding assistant as of launch.

Windsurf's Cascade agent (the multi-step reasoning engine behind Flows) was demoed in November 2024 completing a 10-file refactor with terminal and browser feedback loops — a live demo that went viral and drove 100,000 sign-ups in 72 hours.

Cohere was co-founded by Aidan Gomez, who was a co-author of the seminal 'Attention Is All You Need' paper that introduced the transformer architecture — the technology underpinning virtually every major LLM today.

Cohere's North Star is private deployment: unlike OpenAI and Anthropic, Cohere actively champions running models on-premises or in a customer's own VPC, making it the only major frontier AI company to treat cloud-API access as secondary to enterprise ownership.

Command R7B (December 2024) is the lowest-cost model in the Command family at $0.0375/1M input and $0.15/1M output — under four cents per million input tokens, making it one of the cheapest production LLM APIs available anywhere.

Cohere raised $500M in a July 2024 Series D at a $5B valuation, with investors including Nvidia, Salesforce, Oracle, and Fujitsu — all cloud and enterprise compute partners, not just financial investors.

The Rerank API bills by the query ($2 per 1,000 queries), not by token, making it one of the few AI APIs with a unit of billing that maps directly to an application event rather than raw compute consumption.

Comet's two products price on completely different value metrics: Opik meters spans (LLM/agent trace units), while Comet MLOps meters training hours and data storage.

Opik's Pro Cloud is $19 flat per month for the whole account (up to 50 team members), whereas Comet MLOps Pro is $19 per user per month — the same headline number, two different denominators.

Every Comet plan — including both free tiers — includes access to the free version of the other product family, since Opik and MLOps run on the same underlying platform.

Creatify launched at a beta-pricing model in 2024 with a $135/mo Business tier; by 2026 the top self-serve seat plan was Pro at $49/mo — the headline went down even as included credits and tooling expanded.

Every modality — video, avatar, voice, image and text — bills from one credit currency, and the same per-action credit table governs both the self-serve Platform and the developer API.

Creatify reported crossing $9M ARR in roughly 18 months and raised a $15.5M Series A in May 2025 (co-led by WndrCo and Kindred Ventures, with Nat Friedman angel-investing) to launch AdMax, its end-to-end AI ad agent.

The Starter plan round-tripped on price: $33/mo in late 2024, cut to $19/mo through mid-2025, then back to $33/mo by 2026.

Crowdin meters two usage levers — 'hosted words' (the volume of source content under management) and 'managers' (admin seats) — while keeping translator seats unlimited on every paid plan, including Free.

The Free plan asks for something back: free-tier users 'donate' their translations to Crowdin's shared translation memory.

Crowdin priced by translation 'strings' from launch through at least August 2019 (Tiny/Small/Medium tiers from $9/mo in 2011); the hosted-words meter that defines pricing today only arrived after the 2020 platform overhaul.

Hosted words are multiplicative: a 500-word file pushed to 10 target languages consumes 5,000 hosted words (500 × 10), so a multilingual project burns the word allowance far faster than the source word count suggests.

Crowdin was founded in 2009 in Ukraine by Sergiy Dmytryshyn and has stayed largely bootstrapped, reaching roughly $13M revenue with ~120 staff by 2025 without disclosed venture funding.

Cursor now shows two separate usage pools for individual plans: Auto + Composer and API, with the API pool tied to the selected model's API price.

Teams plans add a Cursor Token Rate of $0.25 per million tokens on non-Auto agent requests, while Auto stays exempt.

Legacy request-based plans still surface a 20% surcharge on Max Mode, which makes the pricing history unusually legible in the docs.

Deepgram prices Speech-to-Text and the Voice Agent API per minute of audio, but Text-to-Speech (Aura) per 1,000 characters and Audio Intelligence per 1,000 tokens — three different metering units on one pricing page.

New accounts start with a $200 free credit and no credit card, then drop straight onto pay-as-you-go rates with no minimums or expiration.

As of May 2026 the pricing page advertises 'limited-time promotional rates on streaming,' showing discounted streaming STT prices struck through against the original rates.

The Voice Agent API offers BYO (bring-your-own) LLM and TTS tiers — e.g. Standard - BYO TTS is $0.065/min vs $0.075/min for the fully managed Standard tier.

DeepInfra publishes a per-million-token rate for nearly every open model it hosts — from Llama-3.1-8B at $0.02 in / $0.05 out to flagship DeepSeek-V4-Pro at $1.30 in / $2.60 out — making it one of the most transparent open-model inference price lists in the market, with prompt-cache rates shown inline.

The same DeepInfra account spans four billing primitives at once: per-token LLM and embedding APIs, per-image Flux generation priced by resolution and step count, per-minute Voxtral audio transcription, and per-GPU-hour on-demand B200/H200 containers — a single bill across four metering units.

DeepInfra raised a $107M Series B to scale its inference cloud, and runs a DeepStart program granting qualifying startups 1,000,000,000 free tokens (valued at DeepSeek-V3.1 prices) for companies that have raised $250K–$10M and were founded within the last two years.

DeepInfra's DeepCluster product flips the usual cloud model: instead of renting capacity, the customer OWNS the NVIDIA B300 hardware (on their balance sheet, eligible for depreciation) while DeepInfra procures, deploys, and operates it — all-in from $1.98/GPU-hr on a 5-year term vs a $6.50/GPU-hr public-cloud reference.

DeepInfra launched in 2023 billing purely by inference execution time ($0.0005/second), only adding per-token LLM pricing in late 2023 — and has since cut its custom-LLM GPU-hour rate roughly 2.5× (A100 $2.00/hr in April 2024 to $0.89/hr by August 2025), a price-cut cadence that made it a go-to "cheap inference" reference in cost-sensitive communities.

DeepInfra was built by the team behind the imo messenger app (200M+ users) and, per its 2026 Series B announcement, processes roughly five trillion tokens per week — 25× the token volume it ran at its Series A.

DeepSeek-R1's January 2025 release caused Nvidia's stock to drop approximately 17% (~$600B in market cap) in a single day — the largest single-day market cap loss attributable to an AI event in history — because R1 demonstrated frontier AI reasoning at roughly 1/30th the inference cost of OpenAI o1.

DeepSeek is funded by High-Flyer Capital Management, a Chinese quantitative hedge fund. DeepSeek reportedly trained V3 on approximately 2,000 Nvidia H800 GPUs at an estimated total cost of $5.5M — a fraction of the 10,000–100,000 GPU clusters used by US frontier labs for comparable models.

DeepSeek-V3 and R1 model weights are open-sourced under the MIT license, allowing any developer or company to self-host the models. This makes DeepSeek the only frontier-class model family that is both commercially cheap via API and fully free for self-hosting.

DeepSeek's cache-hit input pricing is among the most aggressive in the AI API market: V4-Flash cache-hit input is $0.0028/1M — reduced to one-tenth of its launch price in April 2026 — making reused-context input effectively free relative to US frontier models.

DeepSeek-V3 achieved benchmark scores matching or exceeding GPT-4o and Claude 3.5 Sonnet at a training cost orders of magnitude lower, validating the 'mixture-of-experts' architecture as a route to frontier capability at dramatically reduced compute.

Descript was founded in 2017 by Andrew Mason, the former CEO of Groupon, after he spun it out of his audio-tour startup Detour.

OpenAI's Startup Fund led Descript's $50M Series C at a reported ~$550M valuation in November 2022 — making OpenAI both an investor and, via its models, part of Descript's AI stack.

Descript's AI voice-cloning lineage traces to Lyrebird, the synthetic-speech startup it acquired in 2019; the Lyrebird name still appears in Descript's site footer under an ethics statement.

On 23 September 2025 Descript replaced transcription-hour plans with media-minutes plus AI-credit pools — and counts every uploaded file against the media pool, so multitrack uploads burn allowance several times faster than a single mixed file.

Annual billing on Descript buys two things at once: up to 35% off the seat price AND a larger bundled allowance of media hours and AI credits — a packaging lever most seat-priced SaaS tools never pull.

Diffbot's Free plan grants 10,000 credits/month at $0 per credit — no credit card, free forever — enough to extract 10,000 web pages or export 400 Knowledge Graph entity records.

Credit cost differs per activity: extracting a web page is 1 credit, but exporting a single Knowledge Graph entity record costs 25 credits and a summarized facet query costs 100.

The per-credit overage rate falls as you climb tiers — $0.001 on Startup vs $0.0009 on Plus — so higher plans are partly a volume discount on metered usage.

Diffbot offers the entire product suite 100% free to active college/university students aged 16+ via an application-gated academic program.

Diffbot's $299 and $899 anchor prices have not changed since 2016 — through a full migration from per-API-call billing to activity-weighted credits and a Global Index → Knowledge Graph rebrand, the sticker prices held for a decade.

Until 2024 Diffbot's entry plan was a 14-day trial; in April 2024 it became a permanent free-forever plan (10,000 credits/mo, no credit card), flipping the entry tier from trial to true freemium.

Founder/CEO Mike Tung calls Diffbot one of the few bootstrapped, profitable AI companies; it raised a single $10M Series A in 2016 (Tencent + Felicis) and reportedly reached ~$3.1M revenue in 2024 with a ~35-person team.

Dify's Cloud plans meter 'message credits' (Sandbox 200 one-time, Professional 5,000/mo, Team 10,000/mo) rather than raw LLM tokens, so the platform fee is decoupled from the underlying model spend that users pay their own model providers for.

Every paid Cloud tier — even $59 Professional — carries No Dify API rate limit, while the free Sandbox is capped at a 5,000 API-call/month limit, making the API cap one of the clearest free-to-paid upgrade triggers.

Dify is open source under the 'Dify Open Source License' (Apache 2.0 with added conditions); the self-hosted Community Edition is free, and the same product is resold as 'Dify Premium' on the AWS and Azure marketplaces for teams that want custom branding without a Cloud subscription.

Students, teachers and educational staff with a school email can run the Yearly Professional plan for free after annual re-verification — a rare case of a usage-metered AI platform giving its mid-tier away to a whole buyer segment.

Dify (GitHub org langgenius) is one of the most-starred AI repos in the world — roughly 143,000 stars and 22,000+ forks as of mid-2026 — yet its $59 and $159 Cloud headline prices have not moved since the earliest archived pricing page in December 2023.

Digits was founded in 2018 by Jeff Seibert and Wayne Chang, who previously built Crashlytics — acquired by Twitter, then sold to Google and folded into Android/Firebase.

Digits raised roughly $97.5M total, including a $65M Series C in March 2022 led by SoftBank at a $565M valuation, with GV and Benchmark having led earlier rounds.

Digits launched as a free real-time finance dashboard ("Digits for Expenses", unlimited and $0), then repositioned twice — to per-firm accountant tiers, then to AI bookkeeping for businesses and firms with no permanent free tier.

By late 2023 Digits advertised it tracked over $597 billion across 5,000+ businesses; by 2025 its Autonomous General Ledger had powered 2,000+ month-end closes.

In April 2026 Digits introduced outcome-based pricing for its largest firms — billing only when it measurably reduces manual work, after lifting one top-400 firm's automation from ~75% to 98%.

Dropzone AI used to publish a list price: Wayback snapshots show a 'Starting list price' of $24,000/year in mid-2024 that rose to $36,000/year by early 2025, before the company removed all public dollar amounts from the page by 2026.

At the former $36,000/year list price for 4,000 investigations, Dropzone AI worked out to roughly $9 per investigation — a figure third-party guides still cite as 'the most accessible enterprise AI SOC entry point.'

Today the pricing page publishes no dollar prices at all — instead it leads with an interactive ROI calculator that estimates 'At Least $186,000' in annual return from automating 4,000 investigations.

The base plan is metered not by seats or tokens but by investigation capacity: up to 4,000 full alert investigations per year per AI analyst, with unlimited human users included.

There is no free or self-serve tier — every customer, from a three-person SOC to a large MSSP, must request a custom quote.

E2B charges per second of running-sandbox compute, not per sandbox or per API call — the meter only runs while a micro-VM is actually executing, so a paused or stopped sandbox costs nothing in compute.

Upgrading from Hobby to Pro grants zero additional usage credits. The $150/mo Pro fee buys higher limits (longer runtime, more concurrency, bigger machines), not a credit allowance — a deliberate split between the platform-access fee and the metered compute bill.

E2B is operated by FoundryLabs, Inc. and raised a $21M Series A in July 2025 led by Insight Partners (after an $11.5M seed led by Decibel). Its open-source sandbox SDK reports 3M+ monthly downloads and 1B+ started sandboxes, with 94% of Fortune 100 companies cited as users on the enterprise page.

E2B's headline price has not moved across the full Wayback record: $150/mo Pro and the per-second vCPU rates ($0.000014/s at 1 vCPU to $0.000112/s at 8) are identical in the 2024-12 and 2026-05 archived pricing pages — every visible change was packaging, not price.

The default 2-vCPU sandbox bills at $0.000028/second — about $0.10 per hour of running compute before RAM and storage — making short agent runs nearly free while sustained always-on workloads accumulate quickly.

Concurrency is a priced add-on, not just a limit: Pro includes 100 concurrent sandboxes, with Pro+ (600 concurrent) at +$500/mo and Pro++ (1,100 concurrent) at +$1,000/mo stacked on top of the base Pro fee.

ElevenLabs is really three pricing systems in one: a credit ladder, a voice-agent minute meter, and an API meter.

The company renamed Conversational AI to ElevenLabs Agents in 2025, so the billing story is tied to product naming as well as pricing.

Annual billing exists across the subscription tiers, so the monthly pricing page is only the default view, not the whole offer.

Legacy usage-based billing is still documented for older subscriptions, but new subscriptions are pointed to PAYG instead.

ElevenAgents says silence longer than 10 seconds gets a 95% discount, which materially lowers dead-air cost.

Exa prices its API per 1,000 requests rather than per request, so the headline Search $7 actually means about $0.007 per call.

Returning more than 10 results per request triggers a separate per-additional-result charge ($1 per 1k requests on most endpoints), so result count is its own billing dimension.

Exa Agent can bill on auto effort (compute scales to the task) or one of four fixed-effort modes from $0.025 to $2.00 per request for predictable pricing.

Exa launched as Metaphor Systems and sold flat $100 and $250 per-month subscription tiers (Wanderer and Wanderer+) in early 2024 before scrapping them within months for pure pay-as-you-go credits.

Exa's base Search rate sat at $5 per 1,000 requests from January 2025 through early 2026 before rising to $7 in the April 2026 endpoint-card redesign.

fal advertises H100 GPUs 'from as low as $1.89/hr' — a per-second-billed rate ($0.0005/s) that undercuts most on-demand hyperscaler H100 list prices.

fal has no seats, no monthly plans, and no free tier on its public pricing page — every line item is metered per output or per unit of compute time.

fal normalises model-API prices to 'output per $1' on its own pricing page (e.g. 20 seconds of Wan 2.5 video, or 33 Seedream V4 images), turning the price table into a buyer-facing cost comparator.

fal's B200 (184GB) GPU tier is listed as 'contact us' for both per-hour and per-second pricing, gating its newest Blackwell capacity behind sales.

fal says it powers 50% of Poe's image and video generation and low-latency TTS for PlayAI's voice agents — inference volume sold wholesale to other AI products.

fal's pricing page rebuilt itself twice in 18 months: raw per-second unit billing (CPU/GPU/memory) in 2024 became a per-output 'Output per $1' model comparator by mid-2025, tracking its pivot to a generative-media platform.

fal raised four rounds in under three years to a $4.5B valuation (Dec 2025, Sequoia-led) — its $140M Series D roughly tripled the $1.5B mark from its July 2025 Series C just five months earlier.

Fathom's Team plan ($19/user) is priced below its individual Premium plan ($20/user) — the per-user team rate undercuts the solo upgrade once you have 2+ users.

The Free tier is genuinely unlimited: unlimited recordings, transcriptions, and AI call summaries forever, with no meeting or minute cap.

VC- and accelerator-backed startups from 27+ named partners (including Y Combinator and TinySeed) can get up to 2 years of Fathom Team free via the Qualified Portfolio Partner Program.

Firecrawl prices everything in credits but never charges per seat — a 1-developer team and a 50-developer team on the same plan pay the same as long as their page volume matches.

The pricing page geo-detects currency (it rendered in INR with an India flag from some IPs); the USD prices only appear after switching the currency selector.

1 credit = 1 scraped page across Scrape, Crawl, Map, and Monitor — but Search costs 2 credits per 10 results and Interact costs 2 credits per browser minute, so the unit price quietly varies by endpoint.

Credits do not roll over month-to-month except for auto-recharge packs (valid 12 months) and upfront-granted annual Scale/Enterprise credits.

Firecrawl crossed 500K+ signed-up developers and 100K+ GitHub stars before raising much beyond a $16.2M seed/Series A.

Firecrawl's cheapest plan used to be $50/month: the earliest archived pricing page (April 2024, then 'A product by Mendable.ai') had no free tier and sold credit packs at $50, $375, and $1,250 — the entry price only dropped to $0 in the June 2024 rebrand.

Fireworks AI's $7.00/hour H100 (and H200) on-demand price is one of the lowest published rates among managed inference platforms — roughly 30% below Together AI's $5.49–$6.49 H100 dedicated rates only because Together's listed rate excludes Fireworks' full-stack optimization layer.

Fireworks was founded in 2022 by Lin Qiao (ex-Meta), Dmytro Ivchenko, and Pawel Garbacki — Lin Qiao led the PyTorch team at Meta when PyTorch 1.0 shipped, giving Fireworks unusual inference-runtime credibility.

Fireworks' fine-tuning rate card is one of the most granular in the industry: LoRA SFT at $0.50 per 1M training tokens for <16B models, LoRA DPO at $1.00, full-parameter SFT at $1.00, full-parameter DPO at $2.00 — and it scales linearly through the 16B → 300B+ model size tiers.

Cached input tokens get a 50% discount on serverless inference, and batch inference applies the same 50% discount independently — meaning a batched RAG workload with high prefix re-use can land at 25% of the standard input price.

The new $1 trial credit (down from a more generous earlier offer) is one of the smallest in the inference middleware market and suggests Fireworks now relies more on sales-led conversion than self-serve experimentation for revenue growth.

Frase has no free tier — entry is a paid Starter plan at $49/mo (every plan offers a 7-day free trial with no credit card instead).

Every Frase plan, including the cheapest Starter, includes full AI Agent access, API & MCP access, and AI-visibility tracking — plans differ only in volume and team features, not capabilities.

Frase meters AI-search visibility itself: plans cap 'visibility prompts/month' (50 / 200 / 500) and the number of AI platforms tracked (2 / 3 / 5 / 8) across ChatGPT, Perplexity, Claude, and Gemini.

Extra seats are a flat $29/seat add-on on every paid plan above the included headcount (1 / 3 / 5 seats included on Starter / Professional / Scale).

Frase has changed its value metric four times since 2020 — document credits, then articles, then search queries, then SEO documents — before landing on today's articles/audit-pages/visibility-prompts trio.

Frase started as two products in 2020: 'Frase Content' (the SEO optimizer) and an on-site 'Answer Engine' chatbot priced separately at $199.99/mo for 500 answers; the Answer Engine was later dropped from public pricing.

In October 2022, AI-writing company Copysmith acquired both Frase and Rytr and grouped them under a 'Copyrytr' umbrella — yet Frase kept its own brand and continued shipping independent pricing changes.

For a few months in 2025, Frase sold content as pure pay-as-you-go 'Rank-Ready AI Documents' starting at $3.50 each (dropping to $1.67 each at volume) with no subscription required.

Freed's entry Starter plan is the rare consumer-software tier gated by an explicit output cap — 'up to 40 notes per month' — rather than by feature locks alone.

The annual toggle only discounts the top Premier tier ($119/mo monthly drops to $104/mo annually); Starter ($39) and Core ($79) show the same headline price either way, against a struck-through $59 anchor.

Freed sells no free tier — only a 7-day, no-credit-card trial — even though its target buyer is an individual clinician, the most price-sensitive segment in healthcare software.

The Groups plan bundles an organization-wide BAA, SSO, and centralized billing behind 'Talk to sales,' converting a prosumer scribe into a clinic-wide platform sale.

Freed used to charge a single flat $99 per clinician per month; it later unbundled that one plan into the $39 / $79 / $119 ladder, so today's 'unlimited notes' tier (Core, $79) is actually cheaper than the old single price.

Freed scaled to more than 17,000 paying clinicians across 96 specialties and ~$19M ARR before raising a $30M Series A led by Sequoia Capital in March 2025 — growth its investors compared to consumer SaaS like Slack and Zoom.

Freepik bundles ~30+ third-party GenAI models (Google Nano Banana, Sora 2, Kling, Runway, Flux, Seedream, Veo) behind one credit wallet instead of charging per provider.

Annual subscribers get their entire credit allotment upfront and credits stay valid for a full year, while monthly subscribers lose unused credits at each reset.

The Pro tier costs $250/mo but ships 3.6M credits/year — 20% cheaper per credit than lower tiers and includes a 20% extra-credit bonus.

In 2022 Freepik sold a flat stock subscription for ~$8–12/mo with no AI; by 2025 it had rebuilt the page around an AI-credit pool, leaving the $9/$20/$39 core stable into 2026.

Freepik reached roughly $230M ARR (~€196M) with 1M+ paying subscribers and 250+ enterprise clients without ever raising US venture capital, then rebranded the parent to Magnific in April 2026.

Fyxer's annual prices have not moved since at least May 2025: the old Standard plan billed $270/user/yr (= $22.50/mo) and the current Starter still bills $22.50/user/mo annually — the change was packaging, not price.

Fyxer AI Limited was incorporated in October 2023 (UK company number 15189973) and grew ARR from roughly $1M to $17M in under eight months, raising a $30M Series B led by Madrona in September 2025.

Salesforce founder Marc Benioff invested in both Fyxer's $10M Series A (March 2025, led by 20VC) and its $30M Series B (September 2025).

Fyxer charges one seat per user no matter how many inboxes they connect — linking both Gmail and Outlook still counts as a single seat.

Enterprise carries a hard 50-user minimum, so the smallest Enterprise contract is effectively a 50-seat commitment even though the per-seat price is bespoke.

Cisco acquired Galileo: it announced intent on 2026-04-09 and closed the deal on 2026-05-22, after which galileo.ai added a 'now part of CISCO' lockup to its logo.

Galileo had no public mid-tier price until late 2025: through September 2025 the page showed only a free Developer plan (capped at 3 users) and a custom Enterprise plan — the $100/mo Pro tier appeared in the October 2025 redesign.

When Pro launched, Galileo also lifted the free tier's 3-user cap to unlimited users and switched its metered unit framing from 'user-defined metrics' to 'custom evals' plus a 5,000-trace allowance.

Galileo raised a $45M Series B led by Scale Venture Partners on 2024-10-15, bringing total funding to $68M, and cited 834% revenue growth since the start of 2024.

Galileo meters on traces, not seats or tokens: Free includes 5,000 traces/month, Pro 50,000, Enterprise unlimited — both Free and Pro keep users unlimited.

Genspark prices its consumer Plus and Pro plans by credit tier rather than a single dollar figure — Plus starts from 10,000 credits/month and Pro from 125,000 credits/month, and you can switch credit tiers at any time.

Plus and Pro both advertise truly unlimited, zero-credit usage of top-tier chat and image models (the latest SOTA lineup) — but only guaranteed through December 31, 2026.

The Team plan flips the credit-tier model into a flat $30/seat/month with a fixed 12,000 credits per seat; Enterprise doubles per-seat credits to 25,000 but is contract-only at 151+ seats.

Genspark spent its first nine months (from June 2024) as a free AI search engine with no paid plans at all — credits only arrived after it pivoted to the Super Agent on April 2, 2025, which then hit $10M ARR in nine days, faster than Cursor or Lovable.

Genspark quietly removed its consumer one-time credit top-up packs (10,000 credits for $20, 3-month validity) around March 2026 with no announcement; running out of credits now forces a permanent subscription-tier upgrade or a wait for the monthly reset.

GitHub Copilot replaced its premium-request quotas with GitHub AI Credits in June 2026, where 1 AI credit equals exactly $0.01 USD — making the bill a literal dollar-denominated token pass-through.

Code completions and next-edit suggestions are explicitly NOT billed in AI credits and stay unlimited on every paid plan — only chat, agents, code review, CLI, and Spaces draw down credits.

Copilot Business pools its 1,900 AI credits per user at the enterprise level: 100 seats become a shared 190,000-credit pool, so power users borrow from light users automatically.

Annual Pro/Pro+ subscribers who signed up before June 2026 are grandfathered onto the legacy premium-request model (300/1,500 requests, $0.04 overage) instead of the new credit system.

Copilot's individual price sat at exactly $10/month from late 2021 through 2024 — then changed structure four times in twelve months (Free tier, Pro+, premium requests, AI Credits).

GitHub's April 2026 move to usage-based billing drew 767 points on Hacker News, and press reported power users projecting 10×–50× monthly cost increases for heavy agentic workflows.

From 2019 to 2024 Gladly published prices openly on gladly.com — a Customer-Facing/Hero seat ran $150/mo, a Task User $38/mo, and an Admin/Reporting user was free.

When Gladly moved to gladly.ai it deleted its public price table entirely: the new /pricing/ URL now 308-redirects to the homepage, hiding every figure behind sales.

Gladly's 2023 AI repackaging introduced a $0.60-per-assisted-conversation outcome charge for its Sidekick AI — one of the earliest published per-resolution AI rates in CX software.

Gladly raises the seat minimum as you climb tiers: its Superhero package once required a 45-hero minimum versus 10 heroes for the base Hero package.

Gladly was incubated at Greylock and has raised roughly $208M (GGV, NEA, Glynn, Riverwood); founder Joseph Ansanelli previously sold Kana and co-founded Vontu.

Glean's pricing page is the redirect: glean.com/pricing 301s straight to the homepage, so the company has effectively no public pricing surface to archive — Wayback holds no usable pricing snapshots.

Glean tripled ARR from ~$100M to ~$300M in roughly 15 months while keeping pricing entirely gated — a counter-example to the 'transparency drives growth' thesis.

Glean's own docs disclose credit-consumption ranges (e.g. a Thinking Mode premium query ~35-120 FlexCredits) but never the dollar value of a credit — usage mechanics are public, price is not.

Glean shares its name with at least two unrelated products: Meta's open-source code-indexing system (glean.software) and Glean.ai, an AP/finance tool — a recurring entity-disambiguation trap.

Google Gemini 2.5 Flash-Lite outputs tokens at just $0.40/1M — cheaper per output token than any other frontier-grade model from OpenAI or Anthropic as of mid-2026, enabling cost-effective large-scale deployments.

Google's context caching gives a 90% discount on cached input tokens, meaning a developer who sends the same 100K-token system prompt 1,000 times per day saves roughly $11,250/month compared to charging all tokens at standard rate.

The Gemini API free tier via AI Studio requires no credit card, making it one of the most accessible no-commitment AI API tiers in the market — ideal for student developers, hobbyists, and prototype builders worldwide.

Vertex AI's Priority tier charges 1.8× the standard rate for guaranteed capacity — making Google one of the few AI API providers to explicitly price throughput guarantees rather than bundling them into enterprise contracts.

Google introduced regional (non-global) pricing effective July 2026, adding a 10% premium for Gemini 3 models accessed outside global endpoints — the first time Google split Gemini API pricing by geography.

Gorgias bills on tickets, not seats — every plan from Starter to Advanced includes unlimited users, an explicit rejection of the per-agent model Zendesk and Intercom built their businesses on.

Gorgias has priced AI three different ways in four years: a flat monthly Automation add-on (2022), a bundled automation-percentage slider (2025), and finally a transparent flat $0.90 per resolved conversation (2026).

Gorgias's core tier prices have barely moved since 2022 — Basic $60, Pro $360, Advanced $900 are the same numbers Wayback shows on the August 2022 pricing page.

Gorgias's May 2024 round valued it at $530M — a down-round from the $710M it commanded in August 2022, even as ARR grew from $25M to $69M over the same span.

Overage economics reward scale: Starter is +$0.40 per ticket, Basic +$40 per 100 ($0.40/ticket), but Pro and Advanced drop to +$36 per 100 ($0.36/ticket).

Groq's Llama 3.1 8B Instant at 840 tokens/second is one of the fastest published throughput rates for any 8B-class model in commercial inference — the LPU silicon architecture is designed specifically for inference rather than training, which is what makes the speed and pricing combination possible.

Groq was founded in 2016 by Jonathan Ross, the original engineer behind Google's TPU (Tensor Processing Unit) — making it the rare commercial AI inference platform built on bespoke silicon designed by the same engineer who pioneered modern ML accelerators.

Groq's LPU (Language Processing Unit) deliberately avoids the GPU model: each chip has deterministic execution, no HBM (uses on-die SRAM), and no GPU-style branch prediction — the trade-off is lower per-chip memory but vastly higher single-stream throughput.

Built-in tools have explicit per-use pricing: web search at $5–$8 per 1,000 requests, website visits at $1 per 1,000 requests, and code execution at $0.18/hour — making Groq one of the few platforms where agentic tool usage has transparent line-item billing.

Whisper transcription pricing differentiates by model variant: Whisper Large v3 at $0.111 per hour transcribed (higher accuracy, slower) versus Whisper Large v3 Turbo at $0.04 per hour transcribed (lower accuracy, much faster) — the 2.8× price spread reflects the engineering trade-off.

Gumloop's Pro plan has no fixed price — a credit slider sets it anywhere from $37/mo (20k credits) to $1,840/mo (1M credits), then hands off to 'Contact sales' above 1M.

Seats are free on every paid Gumloop plan: Pro includes unlimited seats, so the only thing you pay for is credits — a near-total inversion of the per-seat SaaS norm.

Bringing your own API key cuts agent AI-model credit costs by 50%, and most native workflow nodes (logic, loops, Google Sheets, Slack) cost 0 credits.

Credits don't roll over month-to-month on Pro — only Enterprise plans get rollover — so unused capacity is forfeited each cycle.

Gumloop started life as AgentHub (Y Combinator W24) — its 2024-02-08 Launch HN drew 162 points — and for over a year priced fixed $97 Starter and $297 Pro tiers before scrapping them for today's $37 credit slider.

Harvey publishes no pricing page at all — harvey.ai/pricing returns a 404, and the public site is a pure demo-request funnel with enterprise logos where a rate card would be.

When Artificial Lawyer estimated Harvey's per-seat economics after the LexisNexis deal, Harvey publicly pushed back, calling the assumptions 'wildly off' while still declining to share actual rates.

Harvey crossed $100M ARR in August 2025 — roughly three years after founding — and reported ~$190M+ ARR by early 2026, on its way to an $11B valuation in March 2026.

Co-founder Winston Weinberg is a former litigator at O'Melveny & Myers; that practitioner credibility helped Harvey land 45+ AmLaw 100 firms as customers.

Hedra meters per-second video, per-megapixel images, and per-1,000-character speech all through one fungible credit wallet — a 1-second Sora 2 Pro clip costs 70 credits while a 1-second Hedra Avatar clip costs just 7.

Monthly subscription credits are use-it-or-lose-it and reset each cycle, but separately-purchased credit packs roll over indefinitely and are only spent after the monthly pool is exhausted.

Annual subscribers receive all 12 months of credits upfront at purchase — a 4,000-credit/month annual plan delivers 48,000 credits immediately, to spend at any pace.

Hedra didn't always sell credits: through March 2025 it billed in video-minutes (Basic $10 = 20 min/mo, Creator $25 = 1 hr/mo) with $0.50/minute overage. It switched to a credit wallet within days of its May 2025 a16z Series A.

Hedra's Live Avatars product (launched July 2025) is billed at a flat $0.05/minute of real-time streaming — completely outside the credit subscription — and the company claims it is 15x cheaper than competing streaming-avatar solutions.

Heidi started life as Oscer and was paid-only: a January 2024 Wayback snapshot shows a Clinician license at $199/month ($1,799/year), with no free tier at all. The free-forever scribe arrived around February 2024.

The product's value metric has never been usage. Heidi has only ever charged per clinician (per seat / per FTE) — it explicitly markets 'unlimited consults' and 'unlimited documentation' at every tier, including Free.

Heidi's October 2025 Series B was led by Point72 Private Investments, billionaire Steve Cohen's firm, at a reported ~$465M valuation — taking total funding to roughly US$97M for a company that was the AI-diagnostics startup Oscer just two years earlier.

By the 2025 Series B, Heidi reported working with over 2 million clinicians weekly and processing 70+ million patient visits across 116 countries — but its Trustpilot score had slipped to 3.3/5 across 481 reviews by May 2026, dominated by complaints about lost recordings.

Heidi renamed its paid tiers twice in 18 months: 'Pro/Together' (2024) became evidence-centric 'Evidence Plus / Clinician / Evidence Team / Practice' by late 2025 — moving the upsell story from features to clinical-evidence depth.

HeyGen (founded 2020 by Joshua Xu and Wayne Liang, originally 'Movio/Surreal') only launched its app in September 2022 yet was named G2's #1 Fastest Growing Product of 2025.

HeyGen raised a $60M Series A in June 2024 led by Benchmark at a $500M valuation, after pivoting its cap table away from mainland-China investors.

The Pro plan exposes the same feature set across an 88× credit ladder — 1,000 credits at $49/mo up to 100,000 credits at $4,300/mo — so heavy users scale spend, not capabilities.

HeyGen runs two separate prepaid balances: web-plan 'Premium Credits' (drawn by MCP/OAuth) and an independent API wallet (drawn by Skills/Direct API) — top up one and the other stays empty.

Avatar IV/V videos cost 20 credits/min versus 3 credits/min for Avatar III — a ~6.7× swing that means the engine you pick, not just the runtime, drives your bill.

Higgsfield's live pricing page is a client-rendered SPA that bails out to client-side rendering, so its prices never appear in server HTML, headless capture, or the Wayback archive — the meta keywords still advertise the long-since-renamed 'Basic Pro Ultimate Creator' tiers even though the live cards now read Starter, Plus, Ultra and Business.

Higgsfield renamed and repriced its tiers between January and April 2026: the old Basic/Pro/Ultimate/Creator ladder (roughly $9 to $249) became Free, Starter ($15), Plus ($49), Ultra ($129) and Business ($89/seat) — a credit-metered structure spanning 200 to 3,000+ credits a month.

Higgsfield was founded in 2023 by ex-Snap generative-AI head Alex Mashrabov (who had sold AI Factory to Snap for $166M) and reached a $1.3B valuation on roughly $138M raised, reporting 300,000 paying subscribers and a ~$300M annualized run rate by early 2026.

The platform meters 50+ underlying models — Sora, Veo, Kling, Soul, Nano Banana, Seedance, GPT Image — through one shared 'credit' unit; top-up credits run about $5 per 100, monthly credits do not roll over, and all credits expire after roughly 90 days.

HoneyHive's free Developer tier ships the full observability and evaluation suite — distributed tracing, custom dashboards, dataset curation, annotation queues, and data export are all enabled, not gated behind a paid plan.

Pricing is metered in 'events,' where one event equals a single trace span or metric-label combination sent via OTLP or JSON — total events = trace spans + metrics.

HoneyHive briefly ran a self-serve paid 'Team' plan at 'Starting $99/month' (50K+ events, data exports, unlimited users) in late 2024, then deleted it — collapsing back to a free-plus-Enterprise structure by December 2024.

The free tier's billing meter was rebuilt mid-2024: it started life metered in 'user sessions' and switched to 'events' around June 2024, aligning the meter with OpenTelemetry spans.

HoneyHive offers startup discounts to companies with under $5M of total funding raised, and was co-founded by two Columbia roommates — CEO Mohak Sharma and CTO Dhruv Singh, the latter previously on Microsoft's OpenAI Innovation team.

Ideogram runs two parallel pricing tracks: a freemium consumer subscription priced in credits, and a developer API priced per output image by model and rendering speed.

On the consumer plans, the cheapest model+speed (Upscale 1.0) costs just 0.5 credits per 4 images, while Ideogram 3.0 Quality costs 6 credits per 4 images — a 12x spread inside the same credit pool.

API character-reference calls cost 1.7x–3.3x the base rate: 3.0 Quality jumps from $0.09 to $0.20 per image when a character reference image is included.

Ideogram raised an $80M Series A led by Andreessen Horowitz in February 2024 (≈$96.5M total) on the strength of typography its rivals couldn't match — reviewers peg its in-image text accuracy near 90–95%.

Instantly was bootstrapped to roughly $2.4M ARR in its first 9 months and ~$20M ARR by December 2024, reportedly with no institutional funding — one of the most-cited bootstrapped cold-email success stories.

Instantly started in 2022 as a pure email sequencer that told buyers to get leads elsewhere — its 2022 FAQ literally answered 'Do you also provide leads?' with 'Not yet ;) but there are good options like Apollo.io.' By 2026 it sells its own 450M+ lead database.

Wayback snapshots show Instantly's headline Growth price climbing from $37 (2022) through a yearly-default $30 (2023-2025) to a monthly-default $47 (2026) — the same $47 number that once meant the lead add-on now means the entry email plan.

The annual discount only applies to the Outreach email-sending plans (Growth $47→$37.6/mo); the Instantly Credits lead-database plans show the same price on Monthly and Yearly.

Every Outreach tier includes unlimited email accounts and unlimited warmup — Instantly meters on emails sent and uploaded contacts, not on the number of mailboxes.

Intercom sells Fin AI Agent as a standalone product that runs INSIDE its competitors' helpdesks — Zendesk, Salesforce, etc. — at the same $0.99/resolution with a 50-resolution monthly minimum and no seat fee. It's a rare 'compete on your rival's surface' pricing move that removes switching costs entirely while still collecting outcome revenue.

Intercom's Early Stage Program offers a 93% first-year discount for startups with fewer than 15 employees and under $10M funding, making the Essential plan effectively about $2/seat/month plus 300 free Fin outcomes per month — one of the most aggressive startup discounts in B2B SaaS and a quiet pricing-tier-by-stealth that's invisible to anyone who doesn't qualify.

Fin's $0.99/resolution is one of the cleanest outcome-based meters in AI today — most 'AI product' pricing meters input (tokens, API calls, queries) regardless of whether the user got what they came for. Intercom only bills when Fin closes a conversation without human escalation, which structurally aligns vendor incentives with customer outcomes.

Jasper launched in early 2021 as Conversion.ai, briefly rebranded to Jarvis, then settled on Jasper in 2022 after a trademark conflict with the Iron Man AI character of the same name.

Early Jasper sold capacity in words: the 2022 Starter plan capped output at 20,000 words/month and 'Boss Mode' charged for unlimited generation — the platform has since dropped word caps and shifted entirely to per-seat pricing.

Jasper raised a $125M Series A in October 2022 at a $1.5B valuation, one of the first generative-AI 'unicorns', just weeks before ChatGPT launched and reset the market it had been built on.

Today Jasper publishes only its $59–$69/seat Pro price; even the cost of a second seat is hidden behind 'Contact Sales', so the public floor understates the real cost for any team.

Jasper's marketing page calls Business 'the most popular plan' while only the cheaper Pro plan carries a visible price — the recommended tier is the one you cannot self-serve.

Jina AI collapses its entire retrieval stack — embeddings, reranking, URL-reading, web search, deep-search and classification — onto ONE shared token balance per API key, so the buying decision is 'how many tokens' rather than 'which products'. Most retrieval vendors meter each capability separately.

Jina AI was acquired by Elastic (NYSE: ESTC) in a deal completed October 9, 2025; founder/CEO Han Xiao became Elastic's VP of AI, yet the standalone Jina Search Foundation API and its token-credit pricing continue to operate independently.

Jina's free tier hands every new API key 10 million tokens with no credit card — but they are restricted to non-commercial use under a CC-BY-NC license, an unusually explicit license-gate on a free API allotment.

Jina AI's October 2023 launch of 'the world's first open-source 8K-context text embedding rivaling OpenAI' drew a 563-point, 201-comment Hacker News thread — one of the largest embeddings-model discussions on HN.

A single DeepSearch query can burn roughly 500,000 tokens because the agent iteratively searches, reads full web pages and reasons until it converges — so one 'question' on the shared balance can cost as much as embedding hundreds of thousands of documents.

Juicebox was founded in 2022 by David Paffenholz and Ishan Gupta (22 and 19 at founding) and went through Y Combinator's Summer 2022 batch before launching PeopleGPT in late 2023.

The company crossed $10M+ ARR with over 2,500 customers — including Cognition, Ramp, and Perplexity — while operating with essentially no sales team, a textbook product-led growth motion.

Searches are unlimited on every paid plan; the real value metric is 'contact credits' and 'export credits' — unlocking a profile's emails/phones is what actually meters usage, not the search itself.

Juicebox sells autonomous sourcing 'Agents' as a flat $199/agent/month add-on with unlimited contact and email credits — a seat-independent outcome layer bolted onto a per-seat SaaS.

In September 2025 Juicebox raised a $30M Series A led by Sequoia (Coatue, NFDG, Y Combinator, Lux, BOND participating), bringing total funding to $36M.

Julius is operated by Caesar Labs, Inc. (founded 2022), which raised a $10M seed led by Bessemer in July 2025 — angels included Perplexity's Aravind Srinivas, Vercel's Guillermo Rauch and Twilio's Jeff Lawson.

The free tier's monthly message allowance was quietly cut from 15 messages to 5 between the late-2025 and early-2026 pricing pages — a 67% reduction to the free funnel.

Julius switched its core meter twice: it started metering 'messages per month' (15/250/Unlimited) in 2025 and by mid-2026 had migrated to large annual 'credit' pools (24,000 to 1,440,000 credits/yr) with daily-refresh top-ups.

The current pricing page contradicts itself: the detailed plan cards show annual-billing rates ($16/$37/$375) while a secondary 'Key features' table shows the higher monthly rates ($20/$45/$450) for the same plans.

Julius says 2M+ users have created 10M+ data visualizations on the product.

Krisp's Meeting AI annual prices are exactly half the monthly rate — the 'Save 50%' annual toggle drops Core from $16 to $8/user/mo and Advanced from $30 to $15/user/mo.

Accent conversion is metered by the hour-per-day, not unlimited: Core gets 1 hr/day, Advanced 4 hr/day (speaker side) and unlimited listener side, Enterprise is custom.

Krisp runs three separate pricing surfaces from one product switcher: self-serve Meeting AI seats, sales-led Call Center AI per-agent ($10+/agent/mo), and an application-only Voice AI SDK with no public prices.

Krisp launched in 2019 at $20/mo for a single feature — muting your own outgoing mic noise — with incoming noise removal free. Today's $8 Core seat first appeared in early 2023, after a 2020 dip to $3.33/mo and a 2022 step to $5/mo.

Its 2018 Show-HN-style launch drew a 393-point, 107-comment Hacker News thread; the pandemic then drove 20× user growth, 23× more enterprise accounts and 13× ARR in 2020 alone, earning a spot on TIME's 100 Best Inventions of 2020.

Krisp raised its $14M Series A in two tranches — $5M in 2020 then a $9M extension in Feb 2021 — and says its voice models have processed over 4 trillion minutes of conversation across 200M+ devices.

Kustomer's customer-facing AI agents are billed per engaged conversation ($0.60 each) rather than per seat — an outcome-style meter layered on top of seat-based tiers.

The two current plans (Enterprise and Ultimate) carry no published per-seat price; every headline on the pricing page is 'Talk to Sales' — but Wayback snapshots show the same tiers were openly listed at $89 and $139 per user per month as recently as 2021.

Facebook acquired Kustomer for about $1B in 2020, then spun it back out in May 2023 at a $250M valuation — a roughly 75% write-down — with founder Brad Birnbaum still CEO.

Kustomer still documents three generations of plans in parallel — current Enterprise/Ultimate, legacy seat-based Professional/Business, and legacy usage-based ProfessionalAI/EnterpriseAI/UltimateAI.

Kustomer's per-conversation AI meter is not new: a 2021 snapshot priced 'KIQ for Self-Service' at $0.50 per fully-automated conversation, four years before today's $0.60-per-engaged-conversation AI agents.

Lago's entire managed product is quote-only — neither the Business nor the Enterprise tier shows a single dollar figure on the pricing page; both say 'Contact us'.

The free path is genuinely free: the self-hosted Community edition is open source under AGPLv3 (~9.8k GitHub stars) and runs on your own infrastructure, with no license fee.

Lago processes roughly $829M in invoices monthly and counts Mistral AI, PayPal, Groq, and Synthesia among its customers, despite publishing no public prices.

Lago's growth story began with content, not a launch: a 2022 essay 'Engineers' billing nightmares' hit 204 points on Hacker News before the product was even mature.

Lago pivoted hard — its 2022-01 site sold a data-sync tool with public $50–$1,000/mo tiers; the open-source billing platform that exists today was a clean restart.

Legora was founded as Leya (an earlier name, Judilica, also appears in filings) and rebranded to Legora in February 2025 — a Feb 2025 Wayback snapshot still shows the literal 'Leya|' cursor mid-hero and a founder letter beginning 'We founded Leya.'

Despite raising a $25M Series A, $80M Series B, $150M Series C and a $550M Series D inside roughly 18 months — reaching a $5.55B valuation by March 2026 — Legora has never published a public price or pricing page on either leya.law or legora.com.

Legora's valuation grew from $675M (Series B, May 2025) to $1.8B (Series C, Oct 2025) to $5.55B (Series D, Mar 2026) — an ~8× markup in ten months while its commercial model stayed entirely sales-led.

Legora is built primarily on Anthropic's Claude models and counts elite global firms — Cleary Gottlieb, Bird & Bird, Goodwin, Mannheimer Swartling, Perez-Llorca — among 800 firms across 40+ markets.

lemlist's entry Email plan is $39/mo for UNLIMITED users — a flat-subscription floor that breaks the per-seat norm of cold-email tools; only the Multichannel tier ($109/user/mo) is priced per seat.

Every paid plan bundles lemwarm, lemlist's email-warmup and deliverability engine, at no extra charge — a feature rivals like Instantly and Smartlead historically metered or sold separately.

Enrichment and intent data are pay-per-success: 1 credit = $0.01, charged only when an action succeeds (5 credits per verified email, 20 per phone number, 20–400 per buying-intent signal).

lemlist is built by lempire, a profitable bootstrapped French company (founded 2018 in Paris by Guillaume Moubeche and brothers Vianney & François Lecroart with ~$1,000) that grew to roughly $28M ARR and a reported $150M valuation — set in late 2021 when the founders sold a 20% stake for ~$30M — without raising a venture round.

lemlist's packaging has been rebuilt at least five times since 2020: Silver/Gold/Platinum ($29/$49/$99) → Email warm-up/Outreach/Sales engagement → a four-tier 'find, warm, reach' relaunch with a 450M lead database (2024) → three tiers with an Enterprise plan (mid-2025) → today's flat-fee unlimited-user Email plan plus per-seat Multichannel.

Lightning AI is built by the team behind PyTorch Lightning, the open-source training framework with 350,000+ builders cited on its pricing page.

Every GPU and CPU Studio is billed by the second and drawn down from a credit pool — the headline plan fee mostly buys you a monthly credit allotment, not the compute itself.

The Free tier gives 15 monthly credits that map to roughly 80 GPU hours per month on interruptible (spot) machines — but those free credits expire every month if unused, while purchased credits last 12 months.

Spot/interruptible GPUs are discounted up to 80% versus on-demand, and a single L40S Studio can cost $2.14/GPU/hr on-demand while an interruptible T4 runs as low as $0.52/GPU/hr.

In January 2026, Lightning AI completed a merger with GPU provider Voltage Park, creating a combined company valued at over $2.5 billion with reported ARR above $500 million and a fleet of 36,000+ owned H100/B200/GB300 GPUs.

Linkup gives every new account a $20 credit that auto-refills to $20 every month — a recurring free allowance (~4,000 standard searches) rather than a one-time trial.

Deep search costs 10x a standard search ($0.05 vs $0.005), but the marketing pricing page advertises a Deep range up to $0.55 while the billing docs cap it at $0.055.

Linkup supports the x402 protocol: AI agents can pay per request in USDC on Base with no account or API key, at a flat $0.01 per request.

Until early 2026 Linkup priced in euros with a three-plan layout (Free / Pay as you go / Custom) at €5 and €50 per 1,000 standard/deep queries; the per-request unit price never moved — only the currency, packaging, and free-allowance framing changed.

Linkup is a Paris-founded startup that raised a €3M pre-seed (Seedcamp, Nov 2024) and a $10M seed led by Gradient (March 2026) with angels from Mistral, Datadog, Deel and Dataiku.

LMNT's 2023 pricing page was headlined 'We like to keep it simple' with just three tiers — Free, a single $7/mo Pro, and Enterprise — and the free tier was non-commercial and required attribution to LMNT.

Between late 2024 and early 2025 LMNT cut its per-1,000-character overage rate from $0.17 to $0.05 on Indie (a ~70% reduction) while simultaneously raising the Pro allotment from 500K to 1.25M characters and Premium from 1.5M to 5.7M.

LMNT dropped per-voice-clone fees entirely in 2024 — its 2024-04 page charged $10/mo for each professional voice clone above the included count; by 2024-09 every tier, including Free, advertised unlimited voice clones.

LMNT was founded by Sharvil Nanavati, who previously originated and led the team behind Google Glass; its mid-2023 seed round was led by Elad Gil and Sarah Guo (Conviction).

The Startup Grant gives qualifying sub-20-employee startups 45M credits free over three months (~330 hours of audio per month) with a $0.028/1K overage rate — lower than any published self-serve tier.

Lokalise's headline pricing has more than doubled over its life: the Pro tier went from $160/mo (2019) to $435 (2020) to $585 (early 2022) to $825/mo (late 2022) before the 2025 repackaging retired the Pro name entirely.

The 2025 'new price plans' rebuilt the whole meter — billing moved off hosted keys and all-seat limits onto processed words plus advanced seats, while basic translator/reviewer seats and hosted words became unlimited on every paid plan.

Lokalise charges for two distinct flavors of AI: Standard AI/MT (Google, DeepL) and a premium 'Pro AI' LLM layer with its own annual word pool — Growth gets 50K Pro AI words/yr, Enterprise 400K, each toppable into the millions.

Founded in Riga, Latvia in 2017 by Nick Ustinov and Petr Antropov, Lokalise raised a $50M Series B led by CRV in December 2021; founder-CEO Ustinov handed the CEO seat to Sophie Krishnan in October 2023 and became Chief Innovation Officer.

API usage is hard-capped at 6 requests per second on every plan including Enterprise — a rate limit that does not lift no matter how much a customer pays.

Lorikeet's entry price tripled in under two years: the Start plan was $500/mo in the September 2024 Wayback snapshot and is $1,500/mo by 2026, while Scale doubled from $2,000 to $4,000.

The credit meter was re-architected entirely — early pricing charged by resolution complexity (FAQ-resolved vs complex-workflow tickets), but the 2026 page charges by channel instead (chat/email/SMS vs voice).

Lorikeet quietly dropped self-serve: 2024 plans had 'Get started' online click-through, pay-by-card checkout, but by May 2025 every tier routes to 'Book a demo'.

Founders Steve Hind (CEO, ex-Stripe) and Jamie Hall (CTO, ex-Google Brain) raised a $35M Series A led by QED Investors in August 2025, taking total funding past $50M.

The headline promise is outcome-based: 'We only charge for successfully resolved tickets. If you're unhappy with how Lorikeet handled a ticket, you don't pay for that ticket.'

m3ter is a billing company that publishes none of its own prices — the pricing page describes a four-step custom quote and routes every path to 'Talk to us,' with no dollar amounts anywhere.

m3ter's founders, Griffin Parry and John Griffin, learned usage-based pricing from the inside: their prior startup GameSparks was acquired by Amazon's AWS in 2017, and they spent three-plus years studying AWS's UBP model before starting m3ter in 2020.

m3ter prices its own product on the exact two dimensions it meters for customers — usage data ingested and bill calculations performed — bundled as allowances inside a core platform fee.

Salesforce is a strategic investor in m3ter and picked it as the advanced metering-and-rating engine behind Revenue Cloud Advanced and Agentforce Revenue Management, announced March 4, 2026 — but m3ter stayed independent rather than being acquired.

m3ter's call-to-action has cycled through three go-to-market postures — 'Let's Talk' (2022), a brief self-serve-flavored 'Try m3ter' (2023), and back to fully sales-led 'Schedule a demo' (2024 onward).

Make renamed its core billing unit from 'operations' to 'credits' — for non-AI apps 1 operation still equals exactly 1 credit, so the change is a rebrand of the metric, not a repricing.

Make's pricing page has no plan toggle in the usual sense: a single credit-volume slider (5,000 credits/mo up to 8M+) sets the Make Plan price, which starts at $9/mo for 5k credits.

Make is owned by process-mining company Celonis (the pricing-page footer reads '© 2026 Celonis, Inc.').

Make began life as Integromat, a bootstrapped startup in Prague; Celonis acquired it in October 2020 for a reported $100M+ and rebranded it to Make in February 2022.

After the August 2025 'operations → credits' rename, many AI users reported higher bills: the same workflow now consumes credits dynamically by tokens, file size, pages, or run time instead of a flat 1-op-1-credit rate.

Manus launched invite-only on March 6, 2025; its demo video drew over a million views in 20 hours, and invite codes were resold on Chinese platforms for ¥50,000–¥100,000 (≈US$7,000–$13,800).

Manus prices everything in credits: a single typical agent task consumes roughly 150 credits, so the $20/mo plan's 4,000 monthly credits funds only ~26 typical tasks before add-on packs are needed.

At its March 2025 paid launch the entry plan was $39/mo for 3,900 credits and the top plan $199/mo for 19,900 credits; the current $20 / $40 / $200 ladder (4,000 / 8,000 / 40,000 credits) is a re-cut of that original scheme.

Manus reached $100M ARR by December 2025 — claimed as one of the fastest-ever climbs — built by Singapore-headquartered Butterfly Effect after relocating out of China amid a US CFIUS review of its Benchmark-led round.

Meta agreed to acquire Manus in December 2025 (reported at US$2–3B), but China's NDRC blocked the deal on April 27, 2026; the live pricing page now footers '© 2026 Meta · bringing AI to businesses worldwide.'

Mercor's contractor side advertises an average pay rate of $141/hr (up from $99/hr in April 2026), with individual roles posted at $60–$250/hr — but the buyer-side price (what AI labs pay Mercor) is never shown publicly. Third-party analysts estimate the recruiting fee near 30%.

Every buyer surface — Experts, Data, Enterprise, Partner — terminates in a 'Talk to the team' contact form; there is no self-serve checkout or rate card anywhere on the site, even though the company advertises a $1B+ revenue run rate.

Mercor's three founders — Brendan Foody, Adarsh Hiremath, and Surya Midha — are former Bay Area high-school debate teammates, Thiel Fellows, and (per multiple outlets) among the world's youngest self-made billionaires after the company hit a $10B valuation at age 22.

In March 2026 Mercor was hit by a supply-chain breach traced to the compromised open-source LiteLLM project; the 600-point Hacker News thread and Wired reported a major lab (Meta) paused work in response — a rare trust event for a sales-gated, price-opaque marketplace.

Metronome publishes zero dollar amounts on its pricing page — just a free Starter tier and a 'Talk to an expert' Custom plan, despite being a billing company that sells the ability to meter and price on consumption.

Metronome's free Starter and Custom plan families only appeared between January and February 2026 — for years before that the pricing page was a single 'custom pricing' contact form with no tiers at all.

Stripe completed its acquisition of Metronome on January 14, 2026; the deal was announced December 2, 2025 and press reported a price of roughly $1 billion, though terms were not officially disclosed.

Metronome's metering engine powers usage-based billing for OpenAI, Anthropic, and NVIDIA — three of the largest names in AI.

micro1 used to publish prices: in 2023 it was a developer-staffing marketplace that listed per-engineer hourly rates ($28–$42/hr, averaging ~$34) right on each profile, with a public 'Pricing' nav link — today it publishes nothing and the /pricing path 301-redirects to the homepage.

micro1 grew ARR from roughly $7M at the start of 2025 to about $50M by September 2025, when it raised a $35M Series A at a $500M valuation led by 01 Advisors (Adam Bain and Dick Costolo, ex-Twitter).

Its government page claims 130,000+ vetted candidates across 100+ domains and 60+ languages, and 2,000+ U.S. jobs created in two months.

micro1's research lab frames human data as a '$1 trillion/year market', yet the product itself now carries no published unit rate — the only public billing signal is a 'cost per task' metric mentioned on the government page.

CEO Ali Ansari was 24 when micro1 raised its Series A; board members include DoNotPay founder Joshua Browder and ex-Twitter COO Adam Bain.

Midjourney employs roughly 107 people yet generates an estimated $300–500M in annual revenue — one of the highest revenue-per-employee ratios in AI at ~$3–4M per employee.

Midjourney has never raised external venture capital; founder David Holz bootstrapped the company to profitability from day one, rejecting VC funding even at a ~$10B implied valuation.

The free trial was killed in April 2023 specifically because deepfake images of Donald Trump being arrested and Pope Francis in a Balenciaga puffer jacket went viral — not because of cost pressures.

Midjourney bills in GPU hours rather than image counts: one standard image generation uses roughly 1 minute of GPU time, meaning the $10 Basic plan yields approximately 200 images.

Turbo mode runs at 3.5× Fast speed but consumes 2× the GPU hours — making it cost-equivalent to generating twice as many Fast images.

Mistral's Pro subscription launched at $14.99/mo in February 2025 — a deliberate $5 undercut of ChatGPT Plus ($20/mo).

The same per-million-token meter underpins both the developer API and the consumer Vibe assistant, with Vibe overages billed at API rate via PAYG credits.

Mistral ships open-weight models (Mistral Small 4 under Apache 2.0, Mistral Medium 3.5 under a modified MIT license) and charges per token to call them — a hybrid of open weights plus hosted inference.

Paris-based Mistral raised a €1.7B Series C in September 2025 led by chip-equipment maker ASML, valuing it around $14B.

Mistral renamed Le Chat to Vibe in May 2026, repositioning the chatbot as an autonomous work-and-code agent rather than a chat window.

Modal's per-second billing granularity ($0.001097/sec on H100) means a 5-second cold start costs $0.0055 — among the finest billing granularity in any cloud compute product, two orders of magnitude finer than the per-minute industry standard.

Modal was founded in 2021 by Erik Bernhardsson (ex-Spotify ML, creator of Luigi and Annoy) — making it one of the few infrastructure platforms where the founder is the primary author of widely-deployed open-source data tooling, lending unusual credibility to the developer-experience pitch.

Modal grants up to $10,000 in credits to qualifying startups and academic researchers — one of the most generous credit programs in cloud compute, reflecting the founder's bet on developer-led GTM rather than enterprise sales-led growth.

The Team plan ($250/mo + $100 credits) is the platform's only mid-tier subscription — its presence between the free Starter tier and the quote-based Enterprise tier creates a rare 'committed mid-market' SKU that most serverless GPU competitors omit.

Modal's storage tier ($0.09/GiB-month with 1 TiB free) is positioned to absorb model weights and dataset storage without forcing customers onto S3 — a vertical integration choice that simplifies the developer experience but reduces multi-cloud flexibility.

Motion launched in 2021 as a $19/mo Chrome-extension calendar tool with a tab manager and distraction blocker — its pricing page didn't mention task management or AI at all.

For three months in late 2025 (roughly Sep–Nov), Motion replaced its whole pricing page with an 'AI Employees' grid running from AI Workplace at $29/mo (1,000 credits) up to AI Employees Plus at $599/mo (250,000 credits), then quietly reverted to the simpler Pro AI / Business AI structure by December.

Motion's Series A angel list included Sam Altman, Michael Seibel, and Cyrus Mistry — the early Google Calendar product lead — backing a calendar-automation startup.

Motion's headline marketing claim is 'Finish 137% more work,' and its pricing page links directly to four competitor comparison pages (vs Wrike, Asana, ClickUp, Monday).

Despite a $550M valuation and 100,000+ customers, Motion carries a well-documented Trustpilot pattern of complaints about trial-to-annual auto-conversion charges and refused refunds.

Murf splits its pricing across two motions: the Studio app sells flat-rate subscriptions metered by voice-generation hours, while the Murf API is pure pay-as-you-go at $0.01 per 1000 characters for its Falcon model.

Murf Studio meters voice generation by time per year, not per month — Creator includes 24 hrs/year and Business 96 hrs/year of voice generation rather than a monthly minute bucket.

Murf API credits never expire once purchased, and every API account gets $10 of free credit refreshed every month; early-stage startups can apply for $1,500 in free credits over 3 months.

Across Murf's archived pricing (2021–2024), the Basic ($13→$19) and Pro ($26) Studio tiers barely moved while the Enterprise tier swung from $83 to $166 to a per-seat $59, then $99, then $75 — almost all the price action happened at the enterprise edge.

Murf's pricing page is a JavaScript single-page app, so every Wayback snapshot from September 2024 onward archived as a blank skeleton — making the exact Basic/Pro → Creator/Business rename date impossible to pin from the public archive.

n8n bills by completed workflow executions, not by step, node, task, or user — a single execution can fan out across hundreds of nodes and still counts as one.

Until August 2025 every n8n cloud tier capped how many workflows you could keep active (5 on Starter, 10 on Pro); the 2025 repricing removed that cap and made users, workflows, and steps unlimited on every plan, leaving executions as the only meter.

n8n coined the term 'fair-code' in March 2022 when it swapped Apache 2.0 + Commons Clause for its Sustainable Use License, which it built on top of Elastic License 2.0.

n8n raised a $180M Series C at a $2.5B valuation in October 2025 — led by Accel with NVIDIA's NVentures participating — on the back of 10× revenue growth, after repositioning from 'Zapier alternative' to AI-agent orchestration.

The self-hosted Community Edition is free and open, so n8n competes against its own paid cloud on the same codebase; in 2022 the pricing page led with two free self-host options and put managed Cloud behind a separate early-access tab.

Nomic's main site (www.nomic.ai) pivoted from an embeddings/data-platform company to a vertical AEC (architecture, engineering & construction) agentic platform — its first Wayback pricing snapshot at www.nomic.ai/pricing appeared 2025-11-15, while the legacy Atlas tiers had been archived since October 2024.

The Nomic Platform charges a $1,000/month minimum platform commitment that bundles the first 25 seats — so the practical floor for the AEC product is $12,000/year before any AI usage.

Each $40 seat contributes exactly $20 of AI usage to a single org-wide pool shared across Platform, Assistant, Workflows and the Developer API — half of every seat fee is pre-paid metered consumption.

Nomic Embed text-v1 (released 2024-02-01, Apache 2.0) was the first fully open, reproducible long-context embedding model to beat OpenAI's text-embedding-ada-002 on MTEB — training code and the full 235M-pair dataset were published.

On Atlas, embedding overage is quoted as $1 per 10M text tokens — roughly 100× cheaper than OpenAI's text-embedding-3-small list price per token — and $1 per 50,000 images.

Nooks has never published a dollar price on its pricing page: Wayback snapshots from May 2023 through April 2026 all show named tiers or product cards behind a 'Request Pricing' / 'Get Nooks pricing' button, never a number.

Nooks repackaged its pricing page at least four times in three years — from persona tiers (Cornerstone / Connect / Catalyst, 2023) to Growth/Enterprise (2024) to three 'AI assistants' (early 2025) to today's five product lines — without ever changing the gated, sales-only motion.

The 2023 pricing page offered a 'Try for Free' button next to 'Request Pricing'; by 2024 the free path had disappeared and every CTA routed to sales.

The Contact Data Enrichment package bundles access to six data providers (Wiza, People Data Labs, Apollo, Forager, Datagma, Prospeo) under one Nooks subscription instead of separate contracts.

Nooks raised a $43M Series B led by Kleiner Perkins in October 2024 — the same announcement that rebranded the product an 'AI Sales Assistant Platform' — bringing total funding to about $70M on top of a $22M Series A six months earlier.

Novita publishes per-second billing for both GPU instances and agent sandboxes — a 5-minute coding-agent task on 1 vCPU + 512 MiB RAM is quoted at roughly $0.0034.

The same NVIDIA H100 appears at three different prices depending on product: $2.59/hr on-demand GPU instance, $1.99/GPU-hour as a dedicated endpoint, and $1.70/GPU/hr on an 8-GPU bare-metal node.

Novita lists 226 models on its catalog and undercuts first-party APIs — DeepSeek V3.1 runs $0.27 input / $1 output per million tokens versus DeepSeek's own rates.

Spot GPU instances are priced at roughly half the on-demand rate (RTX 4090 $0.67 on-demand vs $0.34 spot).

Novita started in 2023–2024 as a credit-funded Stable-Diffusion image API billed in USDT/Stripe top-ups ('1/10 the price of DALL-E2 and MJ') with a Singapore HQ — only pivoting into LLM + GPU inference and relisting in San Francisco through 2025.

Only one of Numeric's three tiers carries a public price — Essentials at $30/user/mo; Growth and Enterprise are both 'Custom' and gated behind a demo form.

The entry tier used to be free: 2024–early-2025 Wayback snapshots show a 'Starter' plan reading 'Begin for free, then starts at $30/month/user' — by 2026 the free-trial language was dropped and the tier renamed 'Essentials.'

Numeric's tiers are keyed to ERP maturity, not feature counts: Growth targets QuickBooks/Xero teams, Enterprise targets NetSuite teams — the demo form's General Ledger dropdown effectively routes the quote.

Numeric raised a $51M Series B led by IVP in November 2025 (total funding ~$89M) and launched its Cash Management product the same day — Brex's pilot lifted its transaction match rate from 30% to over 90%.

Angel backers include Marc Huffman (former CEO of BlackLine) and Ron Gill (former CFO of NetSuite) — two of the incumbents Numeric competes against on close management.

GPT-4's launch in March 2023 at $60 per million output tokens made it the most expensive widely-available model in history — within 26 months OpenAI had cut equivalent capability cost by 98% with GPT-4.1 at $8/1M output.

ChatGPT reached 1 million users in 5 days after launch in November 2022 — the fastest consumer product adoption ever recorded at that time. It passed 100M users in 60 days.

OpenAI's $200/month ChatGPT Pro plan, launched December 2024, gives unlimited access to o1 Pro mode — a configuration that uses significantly more compute per query than the standard o1 model and was not previously available at any price.

OpenAI uses a 'soft limit' system for API usage: there is no hard cap by default, but users can set monthly spend limits in the dashboard to prevent runaway costs from agentic loops.

The GPT-4o mini model at $0.15/1M input tokens is 97× cheaper than the original GPT-4 launch price ($15/1M input), while scoring competitively with GPT-4 on many benchmarks — the fastest cost-performance improvement in AI model history.

OpenMeter's pricing page no longer shows any prices — it is now purely a migration notice announcing that OpenMeter Cloud has become Kong Metering & Billing after Kong acquired the company on September 3, 2025.

A billing-infrastructure vendor changed its own pricing model three times in two years: usage-based per-event ($30+) in 2023, flat $249–$349/mo fixed in 2024–early 2025, then back to usage-based ($249/mo + events + a 0.4% billing-volume fee) by mid-2025.

OpenMeter is Apache-2.0 open-source and fully self-hostable via Docker Compose or Kubernetes Helm charts, so the core platform can be run for free indefinitely; the repos moved under Kong but stay open source.

OpenMeter is a Y Combinator W23 company that raised a $3M seed round in March 2024 from Y Combinator, Haystack, and Sunflower Capital before being acquired by Kong about 18 months later.

The legal operating entity behind OpenMeter is Tailfin Cloud Inc., still in the site footer copyright even after the Kong acquisition.

OpenPipe cut its training rates roughly 6x in early 2025: the 8B-and-smaller tier fell from $3.00 to $0.48 per 1M tokens and the 70B+ tier from $16.00 to $2.90, captured between the January and April 2025 docs snapshots.

CoreWeave agreed to acquire OpenPipe on September 3, 2025 (terms undisclosed) to fold its reinforcement-learning agent-training stack into CoreWeave's AI cloud — yet the published rate card was unchanged a month later in the October 2025 docs snapshot.

OpenPipe's launch on Hacker News in September 2023 — 'Fine-tune your own Llama 2 to replace GPT-3.5/4' — drew 955 points and 181 comments, one of the higher-scoring fine-tuning threads of that year.

OpenPipe's open-source ART (Agent Reinforcement Trainer) toolkit has roughly 9k GitHub stars and is the RL framework CoreWeave cited as the strategic reason for the acquisition.

Third-party fine-tunes (OpenAI GPT, Google Gemini) run through OpenPipe at the provider's standard rates with zero markup — you are billed directly by OpenAI or Google, and OpenPipe just passes the API calls through.

Orb briefly published a "$1,750/month, billed annually" Core starting price in late 2024 — then deleted every dollar amount and went fully "Custom pricing" by early 2025, a near-textbook reversal of pricing transparency for a company that sells pricing tooling.

Orb changed its own value metric mid-stream: through mid-2024 it billed purely on monthly event volume and explicitly said invoicing was included "without charging a percentage of billings" — then it added billings (a cut of invoice value) as a primary metric, doing the exact thing it had advertised it didn't do.

Orb's pricing page doubles as a teaching artifact: it annotates itself with 19 "Anatomy of a Pricing Page" best practices (why three tiers, why a 'Most Popular' badge, $99-vs-$100 charm pricing) used to design the page.

Orb was founded in 2021 by two ex-Asana engineers and has raised $44M (seed $5.1M, Series A $14M, Series B $25M led by Mayfield in 2024), powering billing for Vercel, Perplexity, Pinecone, Replit, and Supabase.

Orb's ingestion is stress-tested to volumes such as 250,000+ events per second, and it meters on billings (total invoice value) — so its own bill compounds with its customers' revenue, not just their raw usage.

Oxylabs prices every proxy line on a meter that fits the product: residential and mobile proxies bill per GB of traffic, while ISP, datacenter and dedicated-datacenter proxies bill per IP — so the same vendor runs two completely different value metrics side by side.

Its Web Scraper API uses success-based billing: requests that return 5xx/6xx system errors are not charged, and per-1K-results rates even vary by target (Amazon $0.50, Google $1.00, other $1.15 per 1K without JS rendering).

The cheapest residential entry plan ($30/mo Starter at $6/GB) and the cheapest datacenter pay-per-traffic plan ($11.80/mo at $0.59/GB) differ roughly 10x in per-GB price — a clean illustration of how IP type, not just volume, drives proxy pricing.

Oxylabs' residential entry rate has roughly halved since 2022: Wayback shows its premium 'Next-Gen' residential line at $12/GB in late 2022 and $8/GB in 2024, versus $6/GB in 2026 — a textbook case of falling residential-proxy unit prices across the industry.

Oxylabs is a Tesonet-incubated company — the same Vilnius business builder behind Nord Security (NordVPN), Hostinger and Decodo — and has spent years invalidating rival Bright Data's (formerly Luminati) proxy patents, with a U.S. Federal Circuit affirmation on 2025-08-01 bringing the count to twelve invalidated patents.

Parloa publishes no price table anywhere on its site — even the /pricing path 404s. Multiple third-party reviews put the minimum enterprise contract at roughly $300,000 per year, with average contract value above $350,000, making it one of the most expensive and most fully gated vendors in the contact-center AI corpus.

Parloa tripled its valuation in eight months: a $120M Series C at a $1B valuation in May 2025, then a $350M Series D at a $3B valuation in January 2026, on $50M+ ARR and revenue that quadrupled since its 2024 Series B.

Parloa started in 2018 as a 'Conversational AI' phonebot/chatbot platform — its 2021 site even advertised Alexa Skills and Google Actions — before rebranding the product into the 'AI Agent Management Platform' (AMP) for agentic contact centers.

SAP took a strategic investment and integration stake in Parloa, and the customer roster spans Decathlon, Swiss Life, TUI, KPMG, Allianz, Booking.com, and HealthEquity — a signal the model is built for large-account, sales-led contracts rather than self-serve.

Patronus AI's very first pricing page (Wayback, Sept 2023) was an unfinished template: it listed an 'Individual / Base $25-a-month / Enterprise' table priced in generic 'pages' with literal 'Lorem ipsum' body copy — the $25 tier never reflected a real product.

Patronus AI's Developer plan is fully free with no credit card required, but historical data access (Experiments, Logs, Traces) is capped to the last 2 weeks — a soft retention wall rather than a feature gate.

The optional Patronus API is metered per 1,000 calls, and 'large evaluator' calls cost twice as much ($20/1k) as 'small evaluator' calls ($10/1k) — the only place model size shows up directly in the price.

Patronus AI's per-1k API rates ($10 small / $20 large / $10 explanations) have not moved in Wayback snapshots from December 2024 through June 2026 — roughly eighteen months of price stability.

Patronus AI's Lynx hallucination-detection model (July 2024) was open-sourced on Hugging Face and benchmarked as beating GPT-4 on RAG hallucination tasks at a fraction of the size.

Pebblely is bootstrapped out of Singapore (Pebblely Pte Ltd) and publicly crossed US$1M revenue — its pricing page never exposes a sales-led or enterprise tier, only three self-serve subscriptions.

In its 2024 packaging the top Pro plan offered Unlimited images for US$39/mo; by 2026 the same US$39 Pro tier was capped at 500 images/month — a repackaging from unlimited to a hard monthly allotment.

The 2024 pricing page advertised a custom 'we will create a customized AI for your brand' offer starting at US$3,000/month; that line was removed from the public pricing page in the 2026 repackaging.

Pebblely once had a Free tier (40 images/month); by 2026 the free tier was gone, replaced by a US$9 Lite entry plan with only 30 images/month.

Pebblely's pricing page reports 'more than 25,000,000 images generated' — the value metric (images) is also the headline traction metric.

Perplexity's $200/month Max plan matches OpenAI ChatGPT Pro dollar-for-dollar — a deliberate signal that Perplexity considers itself a peer to the market leader, not just a cheaper alternative.

In July 2024, Perplexity launched a publisher revenue-share program after being accused of plagiarism by Forbes, Wired, and others — effectively monetizing the citations that define its product identity.

Perplexity AI's valuation grew roughly 175× in approximately 30 months: from $121M in April 2023 to $21B by early 2026, fueled almost entirely by subscription growth.

The Sonar API launched in January 2025 as a dedicated search-native API, replacing the earlier pplx-api (October 2023) which hosted generic open-source models with no real-time web access.

Enterprise Pro's SCIM provisioning only unlocks at 50+ seats or with at least one Enterprise Max user — making it the rarest automated provisioning gate in the AI-tools category.

Until late 2024, PhotoRoom showed a single flat consumer price — Photoroom Pro at $89.99/year — directly on its pricing page; by January 2025 those dollar figures had been removed and the page now reveals Pro/Max/Ultra prices only inside the app checkout.

PhotoRoom raised a $43M Series B at a $500M valuation in March 2024, led by Balderton Capital with Aglae and Y Combinator participating, reaching roughly $65M ARR at the raise and $94M ARR by the end of 2024 (89% YoY growth).

PhotoRoom's API offers a 100%-money-back guarantee: you only pay for images that meet your quality standards, and it explicitly promises a refund if you find a better photo-editing API.

The API Partner plan drops the per-image rate to $0.01 — half the standard Basic rate — for consumer apps that process 100K+ images per month and display the PhotoRoom logo in their UI (a 100K-image × $0.01 monthly commitment, which PhotoRoom describes as roughly one thousand dollars per month).

PhotoRoom's most-discussed Hacker News post was not about pricing but infrastructure: a 297-point 2024 consumer guide to renting NVIDIA H100 clusters, reflecting its build-our-own-foundation-model strategy (Photoroom Instant Diffusion).

Pixee abandoned a fully public, self-serve price list. Wayback snapshots show a Pro plan that fell from $39 to a $29 'per GitHub Contributor' rate across 2024–2025 before the whole page went quote-only in early 2026.

Its open-source roots are still free: the Codemodder framework (codemodder-java, codemodder-python) ships under AGPL-3.0 on GitHub, and the historical Pixeebot gave unlimited fixes on public repos for $0.

Pixee raised a $15M seed on 2025-05-22 co-led by Decibel and Wing VC, with angels including early GitHub engineer Zach Holman and HackerOne CTO Alex Rice.

It is one of the few security vendors to bill on a resolved vulnerability rather than a seat — the pricing page literally answers 'What is a resolution?' in its FAQ instead of listing a price.

The founders, Arshan Dabirsiaghi and Surag Patel, came out of Contrast Security; Pixee was founded in 2022 and is based in Palo Alto.

Playground's Free tier dropped commercial use between the 2026-01 and 2026-06 /design/pricing snapshots: every snapshot from 2024-09 through 2026-01 told free users they could 'use images commercially,' but the current surface marks Free 'Non-commercial use — no royalty-free license.'

All paid plans draw from one shared monthly credit pool spanning Nano Banana, GPT Image 2 and Seedream — and Nano Banana Pro costs 4 credits per generation versus 1 for every other model, so a Pro Plus user gets 1,000 standard images but only 250 premium ones from the same pool.

Playground ran three different metering models in under two years: per-day image caps on the legacy /pricing surface (10/day free, 200/day Pro), then a rolling-3-hour image window with separate 'GPT-4o edits' counters, then a unified monthly credit pool — all without ever changing the $15 Pro headline price.

Founder Suhail Doshi co-founded Mixpanel at 20 and shut down his prior startup Mighty (a $30/mo cloud browser) in late 2022 to redirect resources into Playground; the company raised a $40M round in 2023 backed by Y Combinator's Garry Tan to 'advance the field of computer graphics.'

Playground briefly sold a Team tier ($30/mo, $25 annual, 2 seats, SSO + org brand kit) from roughly mid-2025 through late 2025, then dropped it in favour of the consumer-only Free / Pro / Pro Plus ladder visible from 2025-12 onward.

Powerdrill runs two separately-priced products on different domains: the credit-based Bloom app at powerdrill.ai/pricing and the feature-quota Chat app at chat.powerdrill.ai/pricing — with different plan names and price points for each.

The Bloom app meters everything in monthly credits (Pro 5,000 / Plus 11,000 / Premium 60,000), and credits from one-time Credit Packages stack on top of plan credits and stay valid for 12 months.

Powerdrill claims 1.5 million+ users and bundles 'Claude Skills' and 'Nano Banana Pro' image/infographic generation directly into its credit-based data-analysis plans.

Until December 2025 the main powerdrill.ai/pricing page was the cheaper feature-quota app (Free / $3.90 / $9.90 / $29.90); the credit-based Bloom app took over that URL in 2026 and the legacy app was moved to the chat.powerdrill.ai subdomain — so the brand 'moved upmarket' without raising prices on its existing low-cost tiers.

Powerdrill's pricing started in early 2024 as a two-tier, GPT-3.5-metered 'messages per month' plan (Free and Plus $9.90) before being rebuilt twice — into a four-tier GPT-4o ladder, then into a feature-quota model — and finally split into two separately-priced products.

PromptLayer meters overage in a single unit it calls a transaction (txn) — requests, agent-node runs, and eval-cell runs all draw from the same per-txn meter.

The per-txn overage rate actually drops as you move up: Pro pays $0.003/txn, Team pays $0.002/txn — volume discounting baked into the tier, not negotiated.

PromptLayer's legal entity is Magniv, Inc.; archived pricing pages carried a 'Copyright 2025 Magniv, Inc.' footer into 2026 before rebranding to a plain 'PromptLayer ©' line.

The whole pricing page is server-rendered as a JavaScript app — the Wayback Machine never archived promptlayer.com/pricing until December 2025, even though the company shipped its open-source prompt-logging library back in 2022.

PromptLayer raised a $4.8M seed in February 2025 whose angel list reads like an AI who's-who: OpenAI's Romain Huet, Google AI Studio's Logan Kilpatrick, and DoNotPay's Joshua Browder all wrote checks.

Puzzle's free tier has migrated its trigger three times: a flat free-while-in-beta plan in 2023, a $15k monthly-expense cap in 2024, an annual-expenses-under-$25k gate in early 2025, and a $20k transaction-volume threshold by 2026.

Founder Sasha Orloff previously co-founded and led consumer-lending startup LendUp; Puzzle is his second venture, building an AI-native general ledger pitched as a QuickBooks replacement.

Puzzle raised $30M in November 2023 (led by S32 and XYZ Capital), bringing total funding to roughly $50M, before any AI-credit pricing existed on the page.

In 2024 Puzzle briefly metered accrual automation at $0.25 per automation — an explicit per-action usage fee that has since been folded into the AI-credit pool model.

Puzzle includes unlimited users on its Complete and Scale tiers, a direct jab at QuickBooks Online's per-seat add-on charges.

Qodo was called CodiumAI until 30 September 2024, when it rebranded alongside a $40M Series A — the new name avoided constant confusion with VSCodium, the open-source VS Code fork.

Qodo's Teams seat price has whipsawed since the rebrand: $19/user/mo (Sep 2024) → $15 (early 2025) → $30 (mid 2025) — it doubled the same quarter it introduced credit metering.

Until mid-2025 Qodo advertised 'no restrictions on the number of calls/tokens or repositories'; it now meters every LLM request as a credit, with Claude Opus costing 5 credits and Grok 4 costing 4 credits per request.

Qodo's open-source roots run deep: its AlphaCodium paper (Jan 2024) beat Google DeepMind's AlphaCode on CodeContests, and Qodo Merge began life as the open-source PR-Agent.

Credits reset 30 days from a user's first message — not on a calendar boundary — so two developers on the same plan can hit their limit on different days.

Reclaim.ai launched on Hacker News in July 2020 as an 'adaptive calendar app' and pulled 106 points on its Show HN debut — two years before it charged a cent.

Dropbox acquired Reclaim.ai in August 2024 for a reported $40.2 million, folding its 22-person Portland team into Dropbox's productivity portfolio while keeping the brand and pricing page standalone.

Reclaim's entire 2021 paid pricing page literally said 'Free through 2021' — the company ran an uncapped free beta for over a year before launching tiers in 2022.

When rival Clockwise shut down on March 27, 2026, it recommended customers migrate to Reclaim — which offered a 100% price-match guarantee on the next 12-month term through June 30, 2026.

Reclaim brands its automation as 'AI Agents' and meters them by tier: 5 on free Lite, 10 on Starter, 50 on Business, unlimited on Enterprise.

Recraft's API and Studio use two completely different units: the Studio meters in 'credits' (1-2 per image) while the API meters in 'API units' priced at a flat $1 = 1,000 units.

The same image can cost wildly different amounts by model: a Recraft V2 raster render is $0.022 via API while a V4.1 Pro render is $0.25 - an 11x spread inside one product.

Recraft's Free plan keeps the copyright: images generated on the $0 tier are owned by Recraft, public in the community gallery, and carry no commercial rights until you pay.

Studio credits do not roll over month to month, but separately-purchased top-up credits never expire - two opposite expiry policies inside the same wallet.

Regie.ai has changed its core pricing metric three times in four years: per-seat content credits (2022, Pro at $29/mo), then $50,000 per AI-Agent use-case (2024), then back to per-rep packages 'starting at $35K/yr' (2025), and finally published per-user seats ($180/$499) in 2026.

In 2024 Regie.ai explicitly rejected per-user pricing — 'Instead of a per-user basis, we base our pricing on the specific use-case of each Agent' — then reversed course and re-adopted per-seat pricing by 2026.

Regie.ai publishes exact seat prices ($180 and $499/user/mo) but still routes every purchase through 'Talk to sales' on an annual contract.

The Force Multiplier Rep seat bundles 120,000 AI/enrichment credits a year (1,000 accounts / 4,000 contacts) before any Data Package add-on.

The AI Dialer can't be bought on its own — it's only sold packaged with Agents, because its leads are warmed up by the agents.

Relevance AI splits its credit model in two: 'Actions' (a flat charge each time a tool runs) and 'Vendor Credits' (the raw AI-model cost), and it passes Vendor Credits through at wholesale with zero markup.

Vendor Credits roll over indefinitely while you stay subscribed — both the bundled allowance and any top-ups — a rare 'use-it-whenever' stance in usage-based pricing.

You can bring your own LLM API keys on any paid plan to bypass Vendor Credits entirely, so the platform effectively lets you opt out of one of its two metered dimensions.

On 8 September 2025 Relevance AI sunset its Business plan, collapsing the self-serve ladder to Free / Pro / Team and pushing larger buyers to Enterprise.

Relevance AI started life (2023 and earlier) as a 'bring your data to life' analysis and visualization tool — vector search, AI clustering, Tableau-like charts — before pivoting to AI agents and the 'AI Workforce' through 2024.

Its old credit model used to charge LLM cost plus a 20% markup if you didn't supply your own API key; the September 2025 repackaging dropped the markup entirely and made model cost a zero-margin pass-through.

Replicate's per-second public-model billing means a 4-second FLUX Dev image generation on an A100 costs roughly $0.0056 — finer granularity than competitors' per-image flat rates, though the per-image SKU ($0.025 for FLUX Dev) is still published as a simpler alternative.

Replicate was founded in 2019 by Ben Firshman (creator of Docker Compose at Docker) and Andreas Jansson — making it the rare AI infrastructure platform where the founder co-created the single most-used developer tool in containers.

Cog, Replicate's open-source model-packaging framework, predates the company's commercial inference platform by two years — and remains the de-facto standard for packaging ML models with PyTorch and TensorFlow runtimes, similar to how Truss became Baseten's developer wedge.

Replicate hosts over 50,000 public models — by far the largest public-model catalog among managed-inference platforms. The community model directory makes Replicate the canonical entry point for 'is there an open-source model for X' for AI engineers.

Replicate's $1,525/hour H100 dedicated rate ($0.001525/sec × 3,600) is roughly comparable to Modal's $3,949/hour H100 ($0.001097/sec × 3,600) — both substantially undercutting AWS Bedrock and Vertex AI hosted H100 rates while remaining higher than raw AWS on-demand.

Reply.io's Email Volume plan is priced by active-contact volume on a slider (from $49/mo at 1,000 contacts up to higher tiers like $159/mo at 10,000), with unlimited users and unlimited mailboxes — a volume meter rather than a seat fee.

The Multichannel plan bundles email, LinkedIn, calls/SMS, WhatsApp and Zapier-any-channel at a flat $89/user/mo (annual), whereas those same LinkedIn ($69/mo) and Calls & SMS ($29/mo) channels are paid per-account add-ons on the cheaper Email Volume plan.

Reply sells three distinct AI SDR motions: a $500/mo 'Hire Jason AI SDR' agent on the self-serve plans, and a separate $500/mo/client 'Agency AI SDR' tier for agencies running it across multiple client workspaces.

Rev AI quotes its in-house Reverb ASR models per hour ($0.20/hr, Turbo $0.10/hr) but its Whisper-based models per minute ($0.005/min) — two different time units on the same pricing card.

Human transcription is offered through the same API at $1.99 per minute — roughly 600x the per-minute cost of the cheapest Whisper machine model ($0.003/min equivalent).

The free tier is denominated in Reverb ASR time: free credits 'equivalent to 5 hours of Reverb ASR,' redeemable across every Rev AI product.

Audio-intelligence add-ons mix units: Language Translation and Summarization bill per minute, but Sentiment Analysis and Topic Extraction bill per 10 words.

Rev AI's machine-ASR rate fell ~10x from a flat $0.035/min in 2019 to an effective ~$0.0033/min ($0.20/hr) after the October 2024 Reverb launch — while human transcription moved the other way, from $1.50 to $1.99/min.

The October 2024 Reverb relaunch made the headline number bigger ($0.02/min became $0.20/hr) even though the price was cut ~83%, purely because Rev switched the billing unit from minutes to hours.

Rev open-sourced the Reverb ASR + diarization models (arXiv 2410.03930) trained on 200,000+ hours of its own human-transcribed audio, then priced its hosted API at the identical $0.20/hr — with a non-commercial license steering commercial users back to the paid API.

Robin AI once published prices openly: a Wayback snapshot from August 2024 shows a Free Forever tier, a Pro plan at $100/user/month (up to 5 users, 3 reports/month) and a 'Talk to sales' Enterprise tier — all of which had vanished behind a demo wall by 2026.

Robin's public /pricing page stopped resolving between January and April 2025: the last archived snapshot (16 Jan 2025) still listed the $100 Pro plan, but by 16 April 2025 the URL 301-redirected to the homepage. The live /pricing path returns a 404 as of June 2026.

Robin AI was co-founded in 2019 by Richard Robinson, a former Clifford Chance lawyer, and James Clough, a machine-learning scientist — and its earliest product was a Microsoft Word add-in that claimed to cut contract-review time by roughly 85%.

Despite raising ~$43M (Series A $10.5M, Feb 2023, Plural; Series B $26M, Jan 2024, Temasek), Robin AI entered a distressed sale in late 2025: Scissero bought its ~70-person managed-services arm in December 2025 and Microsoft acqui-hired its engineering team in January 2026.

Roboflow denominates almost every billable action — image storage, AI labeling, GPU training minutes, CPU/GPU inference hours, and even third-party LLM tokens — in a single unified "credit," so 1 credit buys 30 minutes of GPU training or 1,000 hosted-API inferences depending on what you spend it on.

The free Public plan hands every user roughly $60/mo of free credits, but the catch is that all datasets and trained models are published openly on Roboflow Universe — privacy starts at the $79/mo Core tier.

Roboflow's credit menu prices frontier LLMs directly: as of March 2026, 1 credit buys 400,000 Claude Opus 4.6 input tokens or 1,600,000 GPT-5.1 input tokens, exposing each model's relative cost inside the same currency as GPU training.

Roboflow has rebuilt its pricing model three times: it billed a flat $0.01/image in 2020, then pivoted to a Professional subscription "Starting at $999/month" in 2021, before abandoning both for the unified credit it uses today — a rare case of a company changing its core billing unit twice in five years.

Rows launched on Hacker News on 2021-11-10 to 202 points and 75 comments — one of the highest-engagement Show-HN spreadsheet launches that year.

Rows changed its value metric twice: per-workspace flat tiers metered by integration tasks (2020–2023), then per-seat AI tiers metered by AI Tasks (2024–2026).

In its 2022–2023 era, Pro cost a flat $249/month per workspace with unlimited members; by 2026 the comparable Pro was $79/month plus $8 per user — a complete repricing around seats and AI.

Superhuman (the renamed Grammarly) acquired Rows on 2026-02-22; rows.com fully winds down on 2026-05-31 after nine years and 2.2 million users, with the team folding into Coda.

Rows reported users ran 17 billion spreadsheet functions and 800,000 AI Analyst prompts over its lifetime, per its own farewell post.

Rox meters its AI sales agents in 'Agent Actions' — a single usage unit whose cost per task scales with the task's complexity and the AI model used, not a flat per-task fee.

Every Rox plan, including the free Starter tier, gets full feature access; the only thing you buy by upgrading is more Agent Actions, not unlocked features.

Unused Agent Actions never roll over — they expire at each cycle reset — but Core users can 'early-renew' to reset their allotment before month-end.

At its November 2024 beta, Rox's Core plan started at $20/month with per-tier account caps (10 / 100 / unlimited); by 2026 the price was $50/month and the account caps were gone — captured prices verified against Wayback Machine snapshots.

Rox reached a reported ~$1.2B valuation by early 2026 on roughly $8M of projected ARR, backed by a $50M raise from Sequoia, General Catalyst, and GV — its founder previously sold observability startup Pixie to New Relic.

RunPod's $0.69/hour RTX 4090 Secure Cloud rate is among the lowest published GPU rates for a workstation-class card — a deliberate positioning play to capture hobbyist and student workloads that hyperscalers price out of reach.

RunPod was founded in 2022 by Zhen Lu and Pardeep Singh, both ex-cryptocurrency-mining infrastructure operators who pivoted hardware from GPU mining to AI inference as the mining-to-AI transition accelerated through 2022–2023.

RunPod runs two distinct clouds: Secure Cloud (enterprise-grade data centers, redundant infrastructure) and Community Cloud (lower-cost, partner-operated DCs with reduced reliability guarantees) — letting customers pick the price-reliability trade-off explicitly per workload.

RunPod's Serverless billing is per-second with flex worker prices from $0.58/hour to $8.64/hour depending on GPU type — among the most granular Serverless rate ladders, covering small single-card workloads (16GB cards) through frontier Blackwell inference (B200 at $8.64/hour).

Storage tier complexity is notable: container disk vs volume disk (running vs idle) vs network storage (standard vs high-performance, with tiered <1TB and >1TB rates) — five distinct storage SKUs that finance teams must aggregate to forecast total storage spend.

Runway prices the same credit currency in two output languages at once: 625 credits is published as both '25s of Gen-4.5 video' and '78 Gen-4 images', so the metered unit means different things per model.

The 'Unlimited' plan still ships a 2,250 monthly credit allowance — unlimited generation only applies to the slower, lower-priority Explore mode, not full-speed output.

Runway's $15 Standard and $35 Pro price points predate 2026: archived pricing from early 2022 shows the same two figures, when the top tier was 'Pro Plus' at $90 rather than today's Unlimited at $95.

Monthly credits never roll over — they reset within 24 hours of the billing date — so any unused allowance is forfeited each cycle.

The developer API auto-upgrades through five usage tiers purely on cumulative spend: hit $5,000 purchased and you jump to Tier 5, lifting the monthly spend cap to $100,000 and concurrency to 20.

Rytr's Free plan meters on characters — 10,000 per month — while both paid tiers drop the meter entirely and advertise 'unlimited' generations.

The $9 tier used to be a character-metered 'Saver' plan (~50,000 chars/mo in 2023); it was rebranded to the unmetered 'Unlimited' by 2024 — the meter became a name.

Rytr went to market in 2021 with an AppSumo lifetime deal reported at $39 one-time, a bootstrapped two-person team selling lifetime access before its subscription tiers existed.

Copysmith acquired Rytr and Frase on the same day — 6 October 2022 — and folded them into a 'Copyrytr' collective; Rytr kept its own brand and pricing.

In December 2024 the FTC ordered Rytr to stop selling an AI review-and-testimonial generator; one year later, in December 2025, the FTC reopened and set the order aside citing the federal AI Action Plan.

Scalenut's annual plans run a permanent-feeling '60% off + double your limits' promotion — annual Starter shows $24/mo against a struck-through $59/mo monthly rate, and content allotments literally double (e.g. 5→10 GEO articles/mo).

The platform meters by content output, not seats: GEO articles, optimizations, keyword clusters, and content-audit web pages are each capped per tier, while the Professional tier gives unlimited workspaces AND unlimited team members.

Only the top self-serve tier (Professional, $80–$199/mo) unlocks Perplexity tracking; Starter and Plus track AI visibility on ChatGPT and Google AI Overviews only.

Backlinks are sold pay-as-you-use through an in-app marketplace with no published unit price — buyers add vetted publisher listings to a cart and pay per link at checkout, priced by each site's DA/DR/traffic/turnaround.

Scalenut launched in 2021 as a managed content + design marketplace ('services, software & Talent') with human senior editors at $299–$549/mo, then pivoted to self-serve AI SaaS in late 2021 — its current GEO platform is its fourth distinct packaging in five years.

The value metric has been re-cut four times: document/word credits (2021) → AI Words + SEO Reports (2022) → SEO Articles + keyword clusters + audit pages (2023) → tracked AI-visibility prompts (2025-2026).

A discount banner has run continuously since 2022, escalating from '60% off' to '50%', '55%', '70% Black Friday lifetime', and back to '60% off + double your limits' — the struck-through monthly rate is effectively the list price buyers rarely pay.

ScraperAPI prices on a credit multiplier, not flat requests: a plain page costs 1 credit, but JS rendering or premium proxies cost 10, and ultra-premium + render costs 75 credits per request — so a 100,000-credit Hobby plan can be anywhere from 1,333 to 100,000 actual scrapes.

Hard-target domains carry fixed multipliers regardless of plan: Amazon is 5 credits, any Google/Bing SERP is 25 credits, and LinkedIn is 30 credits per request.

Only the four largest plans (Scaling and up) get pay-as-you-go overage; on Hobby, Startup, and Business, running out of credits forces an upgrade rather than metered overage.

ScraperAPI began life as a bootstrapped solo project: founder Daniel Ni (Yale '12, ex-Wall Street developer and author of the TLDR newsletter) grew it to roughly $3M revenue and 10,000 customers with about 7 employees before selling it to SaaS.group in August 2020.

ScraperAPI's whole pricing model changed in mid-2022: Hobby went from $29 for 250,000 'API calls' to $49 for 100,000 'API credits', and the per-request credit multiplier was introduced — the same headline tiers held those exact prices for years afterward.

ScraperAPI acquired Traject Data (Rainforest API, SerpWow — ten SERP and e-commerce data APIs) on 2026-04-30, folding structured SERP and marketplace data into the same credit economy.

SerpApi only bills for fully successful searches — requests that are blocked, error out, or return a CAPTCHA cost nothing, a meter most SERP-scraping rivals don't offer.

Standard plans carry no per-search overage line at all: run out of searches and the plan auto early-renews (re-billing the full price and refreshing the bucket) instead of charging per extra call.

SerpApi bundles a 'U.S. Legal Shield' — assuming legal responsibility for scraping and parsing with up to $2M of coverage — across all current plans, framed on Fair Use and First Amendment grounds.

The public price ladder runs all the way to 'Cloud 54M' at $106,050/month for 54 million searches — one of the most transparent high-volume API price tables on the web.

Google sued SerpApi on 2025-12-19 over scraping Search results; the lawsuit drew a 66-point Hacker News thread and SerpApi filed a motion to dismiss in February 2026.

Shortwave was built by ex-Google/Firebase engineers — co-founders Andrew Lee, Jacob Wenger, and Jonny Dimond — and is widely described as the spiritual successor to Google Inbox, which Google killed in 2019.

The $9 Standard plan that Shortwave launched with in 2022 had nothing to do with AI: per the founders' TechCrunch interview, it was a 'Slack-like' model where you paid only to unlock more than 90 days of email history.

Shortwave's 2022 'Show HN: Shortwave: Enjoy Your Inbox' launch thread hit 224 points and 123 comments — one of the strongest HN debuts for an email client in recent years.

Between 2024 and 2026 the entry paid price climbed from a $7 'Personal' tier to a $14 'Pro' floor, and the lineup grew a $100 'Max' tier — a 14x spread from cheapest to most expensive paid seat.

Shortwave converts AI cost into a fixed daily 'AI usage quota' (More / 2x / 6x) that you raise only by upgrading tiers — there is no per-token overage bill anywhere in the product.

Smartlead's two top tiers (Smart at $144/mo and Prime at $315/mo) include UNLIMITED contact storage — a deliberate inversion of the per-contact pricing that competitors like lemlist and Saleshandy charge on.

The Prime plan advertises 5,694,000 email sends per month — an oddly specific cap that implies a derived send-budget formula rather than a round marketing number.

Smartlead is run by 521 Products Pty Ltd out of Sunny Sydney, Australia, and bills entirely in USD with a flat 17% annual discount applied at checkout rather than shown as a separate annual price.

Agencies bolt white-label client workspaces onto Smartlead at $29 per client per month and dedicated SmartServers at $39 per server — the platform is explicitly engineered as resold infrastructure.

Smartlead is bootstrapped to a reported $20M+ ARR with roughly 135 employees — founder Vaibhav Namburi started it in 2021 after 11 failed products, and took zero venture funding, which is why its pricing optimises for self-serve add-ons over investor-pleasing per-unit metering.

Smartlead's Pro tier carried a $94/month headline from at least 2022 through 2024 before today's $78 — the names and meters (Basic/Popular/Pro → Base/Pro/Smart/Prime; active leads → unlimited contacts) churned far more than the prices did.

Socket counts a 'developer' as anyone who made a commit to a scanned repo in the past 90 days — not every named seat, so dormant contributors do not inflate the bill.

Socket's Team tier has been a price rollercoaster: $40/dev/mo at launch (2022), cut to $10 then to $8 after the a16z Series A, then climbed $8 → $16 → $25 across 2025 as the platform expanded.

The Free tier allows unlimited developers and repos but caps usage at 1,000 scans/month and 3 members — a generous on-ramp for individual maintainers.

Every Socket product (Open Source, Threat Intel, Firewall, ExtensionGuard, Basics SAST/Secrets/Container) can be purchased individually, but all products must sit on the same plan tier.

Socket's research blog is a recurring Hacker News front-page fixture — its Shai-Hulud npm-worm writeup hit 1,233 points in 2025, far outdrawing the company's own product threads.

Speechmatics meters Pro speech-to-text by the second, but quotes prices per hour — billing is rounded to the second based on the per-hour rate.

Enabling 'Model Training' (letting Speechmatics use your anonymized data) earns a 33% usage discount — a data-for-credit trade rather than a cash discount.

The free tier hands every account 2,400 free minutes per month split across real-time and batch speech-to-text, plus 1 million free text-to-speech characters.

Spellbook launched what it called the first generative-AI contract-drafting tool in September 2022, running inside Microsoft Word rather than as a standalone web app.

Spellbook is built by Rally (Rally Inc.), a Canadian company headquartered in Saint John and Toronto; the legal AI product carried the spellbook.legal domain before migrating to spellbook.com.

Spellbook's pricing page has shown no public dollar figures in every Wayback snapshot from June 2025 onward — the only public price the company advertises is a '$99 for the first month' CLE-attendee promo on its /99 landing page.

Spellbook raised a $20M USD Series A (about C$27M) in January 2024 led by Inovia Capital with Thomson Reuters Ventures participating, after roughly 10x revenue growth and ~300% customer growth in the prior seven months.

Its public 'trusted by' counter climbed from 3,000+ legal teams in mid-2025 to 4,400+ by mid-2026 — a rare quantified growth signal on an otherwise price-opaque page.

Suno gates commercial-use rights, not just capacity, behind any paid plan — its help center says paid-plan songs let creators collect 100% of royalties with Suno claiming no share.

The Free plan refreshes 50 credits every day rather than monthly, a deliberate retention nudge that brings casual users back daily while reserving monetization rights for subscribers.

Suno raised $250M at a $2.45B valuation in November 2025 (Menlo Ventures, Nvidia's NVentures) the same week it settled Warner Music's copyright lawsuit and acquired Songkick from Warner.

Premier ($30/mo) is the only tier that includes Suno Studio, a generative-AI digital audio workstation launched September 2025 — a feature gate, not a credit gate, separates it from Pro.

Despite model upgrades from v3.5 through v5.5, Suno's headline tier prices ($10 Pro, $30 Premier) have held since at least May 2024 — value rises via model quality, not price.

Superhuman picked its famous $30/mo flat price by surveying users with the Van Westendorp question 'at what price would it be expensive but you'd still buy it' — the median answer was $30, and they rounded up off a '9' on a pricing expert's advice.

Founder Rahul Vohra ran a 450,000+ person invite-only waitlist and built a now-canonical product-market-fit engine around the metric 'how disappointed would you be if you could no longer use Superhuman' — targeting 40%+ 'very disappointed'.

Superhuman is now 'formerly Grammarly': Grammarly acquired Superhuman in July 2025, then in October 2025 renamed the entire company to Superhuman, folding Grammarly, Coda, Mail, and Go into one Suite.

In August 2024 Superhuman quietly cut its flat $30/user/mo to a tiered model — Starter $25 and Business $33 — gating its newest AI (Auto Drafts, Ask AI) behind the higher Business tier rather than metering it as credits.

Superhuman acquired AI-spreadsheet maker Rows (already in this corpus) in February 2026 and is winding the Rows product down by May 31, 2026, folding its tech into Coda.

Surfer started in 2017 as a Chrome extension side project (Keyword Surfer) by Wrocław brothers Lucjan and Michał Suski, and bootstrapped to roughly $15M ARR with no venture funding and under $1M of marketing spend.

Surfer has renamed its tiers at least four times: Hobby/Basic/Pro/Business (2019) → Lite/Essential/Advanced/Max (2023) → Essential/Scale/Enterprise (2024) → Discovery/Standard/Pro/Peace of Mind (2026).

Surfer's value metric has churned almost as often as its names — from per-day/per-month query quotas (2019) to articles-per-year (2022) to Content Editor credits with per-credit overages (2023) to today's 'Documents' plus AI-prompt allotments (2026).

In October 2025 Surfer was acquired by Positive Group, a French/European marketing-software roll-up — and its 2026 pricing page reframes the entire product around winning visibility in AI search, not just Google rankings.

The Enterprise card on the public pricing grid shows a placeholder '$999/mo, Tailored packages' number, but the actual Enterprise plan is fully sales-led and gated behind a 'Talk to Surfer Expert' form.

Synthesia decouples price from seats entirely — paid plans cap editors and guests rather than charging per seat, with the real value metric being generated video minutes.

Editing a video only bills the new seconds you change: a 3-second tweak to a 1-minute clip costs 3 seconds, not 60.

There is no per-minute overage — exceeding your plan caps usage until renewal, so the bill is bounded by design rather than risk of spend spikes.

Synthesia's enterprise page claims adoption by over 90% of the Fortune 100, leaning on SOC 2, GDPR, and ISO 42001 (an AI-management-system standard) as trust signals.

Annual billing front-loads a much larger credit pool (Starter jumps from 1,200 credits/mo to 14,500 credits/yr) on top of the headline 34% price cut.

Tavily began as GPT Researcher, an open-source project data scientist Rotem Weiss built in 2023 to give LLMs real-time web data before ChatGPT had internet access — the company productized it into a paid search API.

The single most expensive call on Tavily's price list is a Research request with model=pro, which can burn up to 250 credits (~$2.00 at pay-as-you-go) in one request — over 100× the cost of a basic 1-credit search.

In February 2026 AI-cloud company Nebius agreed to acquire Tavily for a reported $275 million, with the price rising to as much as $400 million if milestones are met — roughly 11× the $25M Tavily had raised across its seed and Series A.

Tavily never charges for failed work: a failed URL extraction costs 0 credits, and Extract/Map only bill per 5 or 10 successful results, so the credit meter tracks delivered data rather than attempts.

The enterprise contact form quietly doubles as attribution analytics — by late 2025 its 'How did you hear about us?' menu listed LangChain, LlamaIndex, AWS Bedrock, Cohere and GPT Researcher, mapping the open-source channels that drove its bottom-up growth.

Tavus's 2023 pricing page sold a flat $275/mo 'Intro' plan for 200 personalized marketing videos — by August 2024 the same domain advertised 'Transparent usage-based pricing' with a $0 Free tier and pay-as-you-go video minutes, a full repackaging from per-video to per-minute.

Tavus ships three of its own foundation models inside one billed minute: Raven (perception), Sparrow (turn-taking), and Phoenix (rendering) — so a single 'conversation minute' meters an entire multimodal pipeline, not one model call.

The cheapest paid developer plan rose from $39/mo to $59/mo between February and March 2025 while the page rebranded from 'usage-based pricing' to 'Pricing built to scale' and added a '$12k annual' Enterprise floor.

Tavus runs the same conversation-minute value metric two opposite ways: uncapped pay-as-you-go for developers (CVI) and flat consumer subscriptions for PALs ($0 / $20 / $50), where minutes are a hard allowance, not an overage.

Tavus raised a $40M Series B in November 2025 (led by CRV) to reposition as 'The Human Computing Company' — total funding ~$64M from Sequoia, Scale, Y Combinator, and HubSpot Ventures since its YC Summer 2021 batch.

tl;dv's app-level pricing endpoint (gaia.tldv.io/v1/billing/prices) is geo-IP-bound: the same page renders ₹ in India, € in the EU, and $ in the US with no in-page currency selector to override it — the USD figures here were recovered from Wayback Machine renders archived from a US IP.

tl;dv cut its Business plan twice in under two years: $59/seat/mo (annual) in 2024, down to $35 by Q1 2025, then down to $29 by early 2026 — while Pro held flat at $18 the whole time.

The free 'Forever Free' tier gives unlimited recordings and transcription but meters AI features at '10x meetings with AI notes' and '10x Ask AI queries' — and third-party reviewers report the cap behaves as a lifetime allowance, not a monthly reset.

tl;dv was founded in 2020 in Aachen, Germany and raised a single €4.3M seed round in June 2022 (led by K Fund), then scaled to 1M+ teams largely on freemium product-led growth rather than further venture funding.

Business is the only paid seat tier that advertises 'No usage-based AI fees', while Enterprise explicitly re-introduces 'Optional usage-based pricing' — so usage billing appears, disappears, then reappears as you climb the ladder.

Togai is a billing company that publishes zero dollar amounts on its own pricing page — just a free Starter tier and a 'Get Custom Quote' Enterprise plan — even though its product exists to help other companies meter and price on usage.

Togai's own pricing scales on metered event volume and invoice value, not on seats: every plan advertises unlimited users, and the FAQ explicitly says pricing 'is not dependent on the number of users.'

Togai was acquired by NYSE-listed Zuora roughly two years after founding and about ten months after its $3.1M seed round — the pricing page logo now reads 'Togai by Zuora.'

The free Starter tier ships the entire feature set — metering suite, pricing orchestrator, entitlements, multi-currency, credits, discounts, commitments, flat fees, add-ons, invoicing, and dashboards — with no credit card required.

Togai's pricing page references a 'Standard' plan (in the FAQ and the Enterprise 'Everything in Standard +' badge) that has never had a tier card of its own in any archived snapshot since 2023.

Together AI's $4.99/hr H100 reserved rate (7–30 day reservation) is one of the lowest published rates for any managed Hopper-class GPU — and the $9.65/hr reserved B200 sets a similar floor for Blackwell, both undercutting Fireworks' on-demand rates.

Together was co-founded by Vipul Ved Prakash (ex-Cloudmark, Topsy founder), Ce Zhang (ETH Zurich systems professor), Chris Re (Stanford ML/Snorkel), and Percy Liang (Stanford CRFM director) — making it the rare commercial product where two top academic ML labs are co-architects of the platform.

Together's serverless rate card publishes per-model pricing inline on the pricing page (rare among competitors like Fireworks which route to docs), making per-model side-by-side comparison friction-free.

Code Sandbox ($0.0446/vCPU-hour, $0.0149/GiB-hour) and Code Interpreter ($0.03/session) launched in 2025 as separate metered SKUs for agentic and code-execution workloads — adding non-token billing dimensions to the rate card.

Together raised a $305M Series B in February 2025 led by General Catalyst at a $3.3B valuation, with NVIDIA and Salesforce Ventures participation — making Together the highest-valued specialized-inference cloud at that point.

turbopuffer's $1/PB base query rate is the price for scanning a full petabyte of data per query — the result of a February 2026 cut from $5/PB that lowered query costs for the largest namespaces by up to 94%.

turbopuffer has no free tier at all; unlike most vector databases its cheapest entry point is a $64/month usage minimum on the Launch plan.

turbopuffer reached an estimated $100M annualized run-rate in March 2026 having raised less than $1M in total primary capital, per its CEO — an unusually capital-efficient infrastructure business.

The plan tier you pick (Launch, Scale, Enterprise) changes only your monthly minimum and compliance envelope — Launch and Scale charge identical usage rates; only Enterprise adds a 35% usage premium.

turbopuffer publishes a dated public pricing changelog at /docs/pricing-log, so every historical rate change (e.g. the Sept 2024 copy_from_namespace 50% discount) is recoverable from text.

Twelve Labs prices video indexing by the minute of source video, not by tokens or API calls — a metering unit almost unique in the AI-API corpus.

The pricing page lets you flip every rate between a per-minute and a per-hour display; $0.042/minute is shown as $2.50/hour, the identical underlying price.

The free tier grants a one-time, non-expiring 600 minutes (10 hours) of indexing — deleting videos does not refund the allowance back to your balance.

Until 2025 the page billed indexing per capability — Visual $0.033/min, Text-in-video $0.067/min, Logo $0.10/min — before collapsing all of it into one flat $0.042/min rate.

Segment calls multiply the meter: a 60-second window with 4 segment definitions bills 4 minutes, and omitting a time range bills the full video times the segment count.

Unbabel's core enterprise translation platform has no public price page — the only buying surface is a sales-contact form, and that gate has held since at least 2021 (Wayback shows a 'Pricing' nav item in 2022–2023 that has since disappeared).

Its self-serve sibling Widn.ai (TowerLLM-powered) publishes prices openly: Free, $19/mo, $90/mo, and Custom — the transparent counterpart to the gated core.

Widn.ai carries a banner that the platform will be decommissioned on 27 August 2026 and replaced by GlobalLink Now (TransPerfect).

Unbabel (Y Combinator W14) raised a $60M Series C in 2019 led by Point72 Ventures, then sold to TransPerfect in August 2025 on undisclosed terms after pivoting hard into proprietary translation LLMs.

At Widn.ai's November 2024 launch, Unbabel CEO Vasco Pedro publicly predicted AI could replace human translators within three years — unusual candor from a company built on human-in-the-loop editing.

Upstash pioneered per-request serverless Redis pricing in 2021 — billing $0.2 per 100K commands instead of per provisioned instance, so databases scale to zero and cost nothing when idle.

Production guarantees (SLA, multi-zone HA, SOC-2, encryption at rest) are sold as an unbundled $200/mo 'Prod Pack' add-on on any paid plan, rather than gated behind an enterprise tier.

In March 2025 Upstash replaced its old Redis 'Pro' plans — which started at $280/mo — with new Fixed plans starting at $10/mo, a ~28× drop in the entry flat-rate price.

Upstash deprecated its Serverless Kafka product in September 2024 to concentrate on QStash and the durable Workflow engine built on top of it.

Upstash hit $1M ARR within two years of its $1.9M seed, then raised a $10M Series A led by a16z in February 2024 after an unsolicited approach, serving 85,000 developers.

Vast.ai has no list price for GPUs at all — every $/hr rate is set by the individual host machine and floats with marketplace supply and demand, so the same RTX 4090 can cost $0.40 on one host and $0.88 on another at the same moment.

Interruptible instances are a true second-price auction: you set a bid, and your job runs while you are the highest bidder for that machine — get outbid and your instance is paused, not destroyed.

Vast Serverless carries zero pricing premium — it bills at the exact same per-second rate as renting the underlying GPU directly, with no separate serverless tier or surcharge, recruiting the best price-performance worker from the open marketplace.

Storage keeps billing on stopped instances — Vast charges $/GB/hr for disk for every second an instance exists (any state except offline), so a paused instance with a large disk still accrues cost until you delete it.

There are absolutely no refunds on spent credits; unspent credit-card credits can be refunded via chat, but crypto-purchased credits cannot be refunded at all.

Vectara started life as ZIR AI and was founded by Cloudera co-founder Amr Awadallah with ex-Google engineers Amin Ahmad and Tallat Shafaat; it emerged from stealth as Vectara in October 2022.

Vectara's pricing floor went from $0/month to $100,000/year in roughly three years: an early Growth free tier (15,000 queries/month) gave way to a $100/mo Standard plan in late 2024 and then to $100K/$250K/$500K annual deployment tiers by late 2025.

Its original 2022 free tier sized 50 MB of storage as 'enough to load the entire Harry Potter series 5x' — a consumer-grade framing it abandoned entirely for six-figure enterprise contracts.

Vectara now publishes hard annual-commitment floors — $100K/yr (SaaS), $250K/yr (VPC), $500K/yr (On-prem) — where most enterprise RAG vendors show only a contact-sales blank.

The deployment model, not a feature list, is the primary price axis: the same platform costs 2.5x more in a customer VPC and 5x more fully on-prem.

Vercel meters across eight distinct billing dimensions on a single Pro plan — more axes than any other PaaS competitor (Netlify uses 4, Cloudflare Workers uses 3).

Fluid Compute charges $0 for I/O wait time — a category-first 'Active CPU' pricing model that can cut function bills by up to 90% for I/O-heavy workloads.

v0 (Vercel's AI UI generator) is billed separately from the core platform, with Free, Team, Business, and Enterprise tiers.

The Pro plan's $20 monthly usage credit absorbs overages in a fixed priority order: bandwidth first, then edge requests, then function invocations — useful trivia for budget tuning.

Voyage AI was founded in 2023 by Stanford CS professor Tengyu Ma with Hong Liu and Kaidi Cao, and raised a $20M Series A led by CRV (with Snowflake and Databricks participating) before MongoDB acquired it for ~$220M in February 2025.

Despite the MongoDB acquisition, Voyage kept its standalone usage-based API live with unchanged public per-token pricing — the deal was structured in three phases, with Phase 1 explicitly preserving voyageai.com, AWS Marketplace, and Azure Marketplace access.

Every Voyage account gets the first 200 million tokens free on current models (50M on the older voyage-*-2 domain models) — a free allowance roughly 100x OpenAI's, and it persists even after a card is added.

Multimodal models bill on two meters at once — $0.12 per 1M text tokens AND $0.60 per 1B pixels — with each image clamped between $0.00003 (50K-pixel floor) and $0.0012 (2M-pixel ceiling), and each video frame billed as an image.

Voyage has no plans or seats at all: throughput is unlocked by cumulative spend, with ≥$100 billed graduating an account to 2x rate limits and ≥$1,000 to 3x — tiers that never downgrade.

WellSaid's plans price by named-user seat, not by audio minutes — but each tier carries a fixed annual download quota (720/year on Creative, 1,300 on Business, 4,300 on Enterprise) that functions as a hidden usage cap.

The 2026 wellsaid.io relaunch nearly halved the headline Creative seat from $99/mo (held since at least 2020) to $50/mo and dropped Business from $199 to $160 — one of the rare blueprint cases of a vendor cutting its own list price.

WellSaid spun out of the Allen Institute for AI (AI2) incubator in 2018 and raised a $10M Series A in July 2021 led by Fuse, with Voyager Capital, Qualcomm Ventures and GoodFriends participating.

The original 2020 page used four different capacity metrics over four years — 'audio clips,' then 'audio files,' then 'downloads,' and finally 'downloads/year' — without ever charging metered overage.

Despite realistic-voice praise on G2, WellSaid carries a 2.4/5 TrustScore on Trustpilot, driven largely by auto-renewal billing complaints (one reviewer reported a surprise $550 renewal after a cancellation 'didn't go through').

Wispr started in 2021 as a neural wristband company trying to translate silent speech into text via a brain-computer interface. After two years of R&D, the founders realised the throwaway dictation app they had built to test the wristband had product-market fit on its own, pivoted in mid-2024, and shipped the Mac app six weeks later — which hit #1 on Product Hunt for both the day and the week of 2024-10-01.

Wispr Flow meters words-per-week per platform, not minutes of audio — Basic gives 2,000 words/week on Mac/Windows, 1,000/week on iPhone, and (during a 2026 promo) unlimited on Android. For comparison, Otter.ai's free tier caps at 300 minutes/month of audio (~30,000 words at conversational speed) but is single-axis; Flow's per-platform split is the only word-quota model in the AI dictation category.

Flow Pro stayed at $12/user/mo from its 2023 launch through at least mid-2025 — multiple third-party trackers report a 2026 increase to $15/mo on monthly billing while keeping the annual-equivalent rate at $12. The annual-billed effective rate has not moved since launch, making this one of the most stable headline prices in the AI-tools category at a time when Cursor, Replit, and Lovable all repriced in 2025.

Writesonic pivoted from a word/credit-metered AI writing tool into a GEO (generative engine optimization) platform — the legacy per-word pricing is gone, replaced by tiers metered on AI-answer tracking, articles, and site audits.

The core metered unit is now 'answers tracked daily' across AI engines: 50/day on Starter, 300/day on Basic, 600/day on Growth — a meter that didn't exist in the old copywriting product.

Writesonic runs a separate agency price list with a 'pitch project' unit: temporary 1–7 day AI-visibility audits for prospects that auto-pause, billed from $200/mo for up to 20 pitch projects.

Writesonic has changed its core value metric three times in four years — generation credits (2022) → words with a Premium/Good/Average/Economy quality switch (late 2022) → credits again (2024) → GEO 'answers tracked' (2025–26) — a rare case of a single product re-pricing through every major AI-content era.

Founded by Samanyou Garg (YC S21, originally 'Magicflow'), Writesonic raised only ~$2.6M seed and then bootstrapped to multi-million ARR, funding the entire GEO pivot without a later round.

You.com once sold a $30/mo Team plan; in September 2025 it replaced Team with a $200/mo 'Max' plan built around unlimited ARI research reports — a ~6.7x reprice of its top consumer tier, recovered from Wayback snapshots of you.com/plans.

Between March and April 2026 You.com cut its API prices sharply: the Contents API dropped from $10 to $1 per 1,000 pages (10x) and Search dropped from $6.25/$8.00 to a flat $5 per 1,000 calls.

You.com was founded in 2021 by Richard Socher and Bryan McCann, both former Salesforce AI researchers; its September 2024 Series B ($50M, led by Georgian) drew investment from NVIDIA, Salesforce Ventures, and even rival search engine DuckDuckGo.

The Research API charges purely by 'research effort' tier, ranging from $12/1k calls (lite) to a Contact-Sales 'Frontier' tier listed at >$2,000/1k calls — a ~167x spread on the same endpoint.

Every You.com API account starts with $100 in free credit and there are no seats, minimums, or platform fees — billing is entirely per-call / per-page.

Zapier prices its core Platform by *tasks* (workflow steps run), not by seats — even the Team plan's 25 included users are bundled into a task-volume price.

A single task-volume slider (100 → 2,000,000 tasks/mo) re-prices every Platform plan live; the headline price you see is the entry 100-task tier.

Zapier meters its two AI products on entirely different units — Agents by 'activities' and Chatbots by the number of chatbots — so one customer can be billed on three dimensions at once.

Zapier reached a $5B valuation in 2021 having raised only ~$1.4M in venture capital, and has been profitable since 2014 — one of the most capital-efficient SaaS companies ever, at roughly $310M ARR by 2024.

In April 2024 Zapier deleted its entry-level Starter plan and upgraded those users to Professional for free, collapsing a five-tier lineup (Free/Starter/Professional/Team/Company) into today's Free/Professional/Team/Enterprise.

Since January 2024 Zapier's built-in tools — Filter, Formatter, Paths, Delay, Looping, Storage, Tables and Interfaces — no longer count as tasks, so only the actions that touch an external app burn the meter.

ZenRows charges nothing for failed or retried scrapes — and HTTP 404 and 410 responses are explicitly counted as successful results, so a page that no longer exists still bills as a success.

A single dollar balance is pooled across the Scraper API, Scraping Browser, and Residential Proxies, so customers never size separate quotas per product.

Protected pages behind anti-bot systems can cost 25x a basic request once JavaScript rendering and premium proxies stack — turning a nominal $0.10 CPM into $2.50 on the same plan.

ZenRows raised roughly 1.1M euro in its first funding round, announced in a banner across its 2023 pricing page.

Its lowest paid tier rose from $49 in 2021–2023 to $69 by 2026, while the entry request allowance grew from 50,000 requests to 250,000 basic results.