Tavus pricing

About

Tavus builds the Conversational Video Interface (CVI) — an end-to-end pipeline for real-time “AI humans”: video avatars (replicas) that see, listen, and respond face-to-face with sub-second latency. The platform bundles its own perception model (Raven), turn-taking model (Sparrow), and rendering model (Phoenix), and exposes them through a whitelabeled API so developers can embed lifelike conversational video into their own products. Tavus markets itself as “The Human Computing Company.”

Tavus serves two distinct audiences from one technology stack. The primary buyer is the developer or team building conversational-video agents — for interviews, coaching, onboarding, customer service, and healthcare — billed through the developer CVI plans. The second is the consumer, served by PALs: AI companions (Noah, Sophie, Charlie, and others) you can text, call, or video-chat, sold as flat monthly subscriptions.

Tavus came out of Y Combinator’s Summer 2021 batch and has raised roughly $64M across four rounds — a $6.1M Sequoia-led seed, an $18M Series A led by Scale Venture Partners (announced March 2024, alongside open developer access to its face-and-voice-cloning API), and a $40M Series B led by CRV in November 2025 with Sequoia, Y Combinator, and HubSpot Ventures participating. The Series B reframed the company as “The Human Computing Company.” Tavus does not publish ARR; third-party estimates put 2023 revenue near $1M, and the company stays private about user and customer counts. The business has materially repositioned since launch — from a personalized-marketing-video tool for sales teams into a real-time conversational-video infrastructure layer for developers, with the consumer-facing PALs companions layered on top in 2025–2026.

Competitively, Tavus sits at the intersection of two markets. Against avatar-video generators like Synthesia, HeyGen, and D-ID, Tavus differentiates on real-time, sub-second conversational video (face-to-face dialogue) rather than pre-rendered talking-head clips — and prices accordingly, on live conversation minutes instead of rendered-video seats. Against conversational-AI platforms it differentiates on owning the full perception-to-render stack (Raven, Sparrow, Phoenix) as first-party models, which is what lets it meter a whole multimodal pipeline as a single billed minute.

Pricing summary : Hybrid CVI access fee plus pay-as-you-go video minutes

Tavus runs two parallel pricing models:

1. Developer CVI plans (hybrid): A monthly access fee (Free / $59 Starter / $397 Growth / Custom Enterprise) bundles a fixed allowance of conversational-video minutes, then bills pay-as-you-go with no overage cap for usage beyond the allowance. The headline metered unit is video minutes — both real-time conversational minutes (CVI) and pre-rendered video-generation minutes — with rates that fall as you move up tiers ($0.37/min → $0.32/min CVI; $1/min → $0.90/min generation). Custom replica training and conversation recordings carry their own per-unit overages.
2. PALs consumer plans (flat subscription): A separate freemium ladder — Free $0, Plus $20/mo, Max $50/mo — that gates unlimited messaging by an allowance of voice & video call minutes (15 / 150 / 500).

What makes this different: Tavus meters the same underlying technology two ways — as a per-minute developer API with uncapped pay-as-you-go usage, and as a flat-rate consumer subscription — so the value metric (conversation minutes) is identical even though the packaging is opposite. The developer side is a textbook hybrid pricing model: a recurring access fee plus usage-based billing on the metered minute, with the included allowance acting as a prepaid bucket before overages kick in.

Pricing by product

Developer CVI (API plans)

Developer CVI — per-unit overage rates (pay-as-you-go)

PALs (consumer plans)

Startups & Grants program

Tavus offers early-stage builders free access for 3 months — 15,000 minutes, 25 replicas, and full developer-platform access (CVI, Knowledge Base with 30ms RAG, Memories). It also runs an academic research grant. There is no list price; access is requested via the Startups & Grants / Enterprise “Talk to us” link. SOC 2, HIPAA, GDPR, and BAA compliance are noted on this surface.

Sales motions across products: PLG / self-serve for the developer Free, Starter, and Growth plans and all PALs consumer plans; sales-led for Enterprise and the Startups & Grants program.

Hidden costs : What heavy conversational-video usage actually costs

The advertised access fee understates what heavy conversational-video usage actually costs once pay-as-you-go minutes stack up. Because Tavus’s overage is uncapped — “never paused, never throttled” — the marginal minute is the real cost driver, and a single popular product feature can drive the metered line far past the base fee.

Archetype A — a Growth-plan team running ~5,000 conversational-video minutes/month. The $397 access fee includes 1,250 CVI minutes, so 3,750 minutes spill to pay-as-you-go at $0.32/min. Add a handful of custom replicas and recordings:

The metered minutes ($1,200) cost roughly 3× the headline access fee ($397) — so at moderate scale the “$397/mo” plan is really a $1,700+/mo bill, and the access fee is a rounding error against usage.

Archetype B — a Starter-plan startup with one viral week (~600 CVI minutes). The $59 fee includes only 100 CVI minutes, so 500 minutes overflow at $0.37/min, plus a custom replica beyond the 3 included:

A 5× jump in usage turns a $59 plan into a $309 bill — so the uncapped overage is the line item early-stage teams must forecast, not the sticker price.

Want to estimate your own Tavus bill? Use the Tavus pricing calculator to model your monthly cost based on conversation minutes, video-generation minutes, and custom replicas.

Pricing evolution : How Tavus’s CVI and PALs pricing has shifted

Cadence

Tracked range: 2023-03–2026-06 (Wayback Machine monthly snapshots of tavus.io/pricing). Quarters not listed were verified stable (0 price changes, 0 SKU additions).

Notable changes

• 2023-03 — Pricing page sells a flat $275/mo “Intro” plan (1 AI voice/face model, 200 personalized videos/mo) — the per-video marketing-automation model.
• 2023-09 — Self-serve plan removed; page shows only an Enterprise “Request Demo” card with no list price.
• 2024-08 — Page relaunches as “Transparent usage-based pricing”: Free $0, Starter $39/mo, Growth $375/mo, Enterprise Custom — all pay-as-you-go on the conversational-video minute.
• 2025-03 — Entry prices raised: Starter $39 → $59, Growth $375 → $397; page rebranded “Pricing built to scale” with a “$12k annual” Enterprise floor.
• 2025-11 — $40M Series B led by CRV; total funding ~$64M; “The Human Computing Company” positioning (VentureBeat).
• 2026-02 — PALs consumer plans (Free / Plus $20 / Max $50) appear; pricing page adds a PALs | Developer tab toggle.

The 2023→2024 repackaging in detail

The most consequential pricing event in Tavus’s history is not a price change but a value-metric change. In early 2023 the company billed per personalized video — the $275/mo Intro plan bought “200 Tavus-generated videos per month,” a packaging built for marketing and sales teams sending one-to-many personalized clips. By August 2024 the same URL billed per conversation minute through a usage-based developer API, with a $0 Free tier inviting developers to “get their feet wet.” That is a complete repositioning of what is being sold (a real-time conversational pipeline, not a rendered video) and who is buying it (developers embedding an API, not marketers buying seats). The per-video plan and the per-minute plan are different products that happened to share a domain — which is why no single SKU shows a clean numeric delta across the boundary.

What’s unique : The distinctive mechanics of Tavus’s two-sided model

1. One value metric, two opposite packagings. Tavus sells the same conversation minute to developers as an uncapped pay-as-you-go API overage and to consumers as a flat-rate allowance inside a PALs subscription. Most companies pick one packaging; Tavus runs both off an identical unit, which keeps the metric honest — the thing customers pay for (a minute of real-time AI-human conversation) is the same thing whether you’re a developer or a consumer.

2. The billed minute is a whole pipeline, not a model call. A single conversation minute meters Tavus’s full first-party stack — Raven (perception), Sparrow (turn-taking), and Phoenix (rendering). Because Tavus owns all three models, it can price the outcome (a lifelike conversational minute) rather than the inputs (tokens, frames, API calls), which is far easier for buyers to reason about than a decomposed multimodal bill — and it sidesteps the value-metric problem that traps vendors trying to bill on raw model usage.

3. Uncapped overage as a deliberate reliability promise. Tavus markets pay-as-you-go that is “never paused, never throttled.” For a real-time conversational product that’s a feature, not a trap — a hard quota cutoff mid-conversation would break the user experience — but it shifts forecasting risk to the customer, which is why a Growth plan can quietly become a multi-thousand-dollar bill (see Hidden costs).

4. Multiple metered SKUs under one access fee. Beyond CVI minutes, Tavus separately bills video-generation minutes ($1 → $0.90/min), custom replica training ($65 → $40/replica), and conversation recordings ($0.03/min) — so the access fee is a gateway to several independent meters, each with its own included allowance and per-tier discount. This is closer to a cloud-infra bill than a SaaS seat plan.

Strengths & weaknesses

Billing UX : Self-serve plan selection and uncapped pay-as-you-go

• In-portal plan selection — plans (Basic / Starter / Growth) are chosen self-serve from the billing page inside the Tavus developer portal; upgrade, downgrade, or cancel at any time.
• Pay-as-you-go with no overage cap — once a plan’s included minutes are used, additional usage is billed at the stated per-minute rate and is “never paused, never throttled” — there is no hard stop or quota lockout.
• Annual billing toggle (20% discount) — choosing annual saves 20% on a paid plan and locks the rate in for the term.
• Conversation transcripts & recordings — full transcripts are exposed on paid plans; conversation recordings are billed as a separate $0.03/min line item on Growth.
• “Talk to us” / Enterprise sales link — Enterprise quotes, Startups & Grants credits, and academic-research access are all requested through the Enterprise “Talk to us” link rather than self-serve checkout.
• Trust Center & compliance surface — SOC 2, HIPAA, GDPR, and BAA compliance status is published via the linked Trust Center.

Strategic wins : Why specific Tavus pricing decisions worked

1. Repricing the unit, not just the number

Tavus’s biggest pricing win was changing what it sells — from a per-video marketing plan to a per-minute conversational API — when the product itself changed. Rather than bolt API pricing onto the old video-quota plan, it relaunched the page around the conversation minute, the unit that actually tracks the value of real-time AI humans. Teams considering a value-metric change should study how cleanly Tavus swapped the metric instead of grandfathering an ill-fitting one.

2. Leading with a $0 Free tier that exposes the whole pipeline

The Free plan gives developers 25 real conversation minutes through the full CVI stack — not a watered-down sandbox — so the “aha” moment (a working AI human) happens before any payment. This is classic product-led growth for an infrastructure product: the cost of a free minute is real compute, but it converts because the demo is the product. The same dynamic — give away the value metric, monetize at scale — drives most usage-based AI pricing launches.

3. Falling per-minute rates that reward growth automatically

By cutting CVI overage from $0.37 to $0.32/min and generation from $1 to $0.90/min as customers move Starter → Growth, Tavus builds a volume discount into the self-serve ladder. Customers see their marginal cost drop as they scale without negotiating, which reduces the incentive to churn to a competitor at the exact moment usage is climbing.

4. Uncapped overage that protects the core experience

Choosing “never paused, never throttled” over a hard quota is a pricing decision that defends the product. For a live conversational interface, a mid-call cutoff would be a worse outcome than an unexpected charge, so Tavus aligned its billing policy with the reliability its category demands — and turned an operational necessity into a marketing line.

Areas to improve : Specific pricing gaps with proposed fixes

1. Add an optional spend cap or budget alert

The uncapped overage is right for reliability but wrong for forecasting — a Growth plan can silently become a $1,700+/mo bill. Tavus should offer an opt-in monthly spend ceiling with email/Slack alerts at 50/80/100% of budget, so finance teams can adopt without fearing a runaway minute. This preserves the no-throttle promise for those who want it while de-risking adoption for the budget-conscious. See our take on the shift from entitlements to credits in LLM billing.

2. Publish a stable “what $397 buys” spec

Included allowances have moved between snapshots (Growth’s bundled minutes shifted materially over 2025), which erodes trust in the headline. Tavus should version and date the included-allowance table on the pricing page so buyers can see exactly what each tier includes today — and what changed — rather than discovering shifts only via the dashboard.

3. Consolidate the four meters into one estimate

CVI minutes, generation minutes, replica trainings, and recordings each bill separately, so no buyer can eyeball their bill. A first-party, interactive estimator on the pricing page (the way we model it in the Tavus pricing calculator) would let prospects price a realistic workload across all four meters before signing up, reducing the “death by a thousand line items” surprise.

Key takeaways

1. When the product changes, change the unit — not just the price. Tavus repriced from per-video to per-minute when it pivoted from marketing video to conversational API. Forcing new value into an old metric (or vice versa) is how pricing pages drift out of sync with what’s actually sold.
2. A real Free tier that runs the whole pipeline converts better than a sandbox. Giving away 25 genuine conversation minutes — not a degraded preview — lets the product sell itself before any payment, which is the PLG engine for an infra company.
3. One value metric can serve two markets. The same conversation minute powers an uncapped developer API and a flat consumer subscription. A clean metric travels across packaging; a packaging-specific metric does not.
4. Uncapped overage is a product decision, not just a billing one. For real-time experiences, throttling at a quota is worse than a surprise charge — so the billing policy should defend the core UX, but pair it with opt-in budget controls.
5. Stack multiple meters carefully — and make them estimable. Tavus bills four separate usage dimensions under one access fee; without a first-party estimator, buyers can’t predict their bill, which slows adoption even when each rate is fair.

UBP implications

1. Outcome-shaped metering is viable when you own the stack. Because Tavus owns Raven, Sparrow, and Phoenix, it prices the outcome (a conversation minute) instead of decomposed inputs. Vendors who own their full pipeline can offer simpler, more defensible usage metrics than those reselling third-party models. It’s the same pattern emerging across AI-agent pricing, where the billed unit increasingly maps to an outcome rather than a raw API call.
2. Hybrid access-fee-plus-usage works best when the included bucket sets expectations. Tavus’s access fee buys a minute allowance that doubles as a forecasting anchor; the overage then scales without a new contract. This hybrid model balances predictable revenue with usage upside — but only if the included allowance is stable and clearly published.
3. The same metric can bridge developer and consumer monetization. Tavus shows that a well-chosen unit lets a company run PLG/API and flat-subscription consumer pricing in parallel without inventing two value stories — a template for AI companies straddling builder and end-user markets.

Sources

• Tavus developer pricing page (accessed 2026-06-01)
• Tavus PALs consumer pricing page (accessed 2026-06-01)
• Tavus Startups & Grants program (accessed 2026-06-01)
• Tavus developer docs (accessed 2026-06-01)
• Tavus pricing page — Wayback Machine snapshot, 2026-05-21 — independent second source (accessed 2026-06-01)

Bottom line

Tavus is a case study in repricing the unit, not the number: it walked away from a flat $275/mo personalized-video plan and rebuilt its pricing around the conversation minute — a single metric clean enough to power both an uncapped developer API and a flat consumer subscription. The model is elegant and honest about value, but the uncapped overage and four parallel meters push real forecasting work onto the buyer, so the headline $59 and $397 plans understate what production usage actually costs. Get the value metric this right, add spend controls, and you have a pricing page that scales as smoothly as the product.

Want to compare Tavus against other conversational-video and avatar pricing? Browse the pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Current developer CVI tiers + PALs (captured)

Jun 2026

Developer plans: Free (25 min), Starter $59/mo (100 min), Growth $397/mo (1,250 min), Enterprise custom — pay-as-you-go video minutes with no cap ($0.37→$0.32/min CVI; $1→$0.90/min generation). PALs consumer: Free, Plus $20/mo, Max $50/mo.

PALs consumer plans added; pricing page splits PALs / Developer

Feb 2026

The pricing page gained a PALs | Developer tab toggle. Developer tiers stayed Basic Free / Starter $59 / Growth $397 / Enterprise Custom; a new consumer ladder (PALs Free / Plus $20 / Max $50) sold the same conversation tech as a flat subscription gated by call minutes (15 / 150 / 500).

$40M Series B; 'Human Computing' repositioning

Nov 2025

Tavus raised a $40M Series B led by CRV (Scale, Sequoia, Y Combinator, HubSpot Ventures), bringing total funding to ~$64M and repositioning as 'The Human Computing Company' — the corporate backdrop to the PALs consumer launch.

Entry prices raised; 'Pricing built to scale'

Mar 2025

Between Feb and Mar 2025 Starter rose $39 → $59/mo and Growth $375 → $397/mo; the page rebranded to 'Pricing built to scale' and added an Enterprise floor 'starting at $12k annual'. Pay-as-you-go-with-no-limits messaging stayed.

Pivot to usage-based CVI API pricing

Aug 2024

Pricing page relaunched as 'Transparent usage-based pricing' with four developer tiers: Free $0, Starter $39/mo (Most popular), Growth $375/mo, Enterprise Custom — all 'pay as you go usage' on top of the access fee. The value metric became the conversational-video minute (CVI), replacing per-video billing.

Pricing page goes sales-only (Enterprise-only)

Sep 2023

By September 2023 the self-serve $275 plan disappeared; the pricing page showed only an Enterprise 'Request Demo' card with no list price, coinciding with the seed round and the shift toward a developer/API platform.

Personalized-video era: flat $275/mo 'Intro' plan

Mar 2023

Tavus's pricing page sold marketing-video automation: an 'Intro' plan at $275/mo (1 AI voice & face model, 200 Tavus-generated videos/mo, unlimited templates) plus a 'Business' Custom plan. Billing was per personalized video, not per minute — the pre-CVI, pre-API model.