Tavily pricing

About

Tavily is a web-search and data-retrieval API built specifically for AI agents and LLM applications. Its endpoints — Search, Extract, Map, Crawl, and Research — turn live web content into clean, agent-ready results, letting developers add real-time web access to an agent in a few lines of SDK code rather than wiring up their own crawlers, proxies, and parsers. The positioning line on the site is “Power your AI with real-time web search,” and the footer claims the product is “Built by researchers. Trusted by 2M+ AI builders.”

Tavily grew bottom-up out of GPT Researcher, an open-source project founder Rotem Weiss built in 2023 to give LLMs real-time web data before ChatGPT had internet access. The company raised a seed round in mid-2024 and a $20M Series A led by Insight Partners and Alpha Wave Global in August 2025 (total funding $25M), reporting over a million developers and more than three million monthly SDK downloads. Named customers include IBM, Cohere, Groq, MongoDB, and Writer. A “Free for students” program offers the service at no cost, reinforcing the research-oriented adoption motion, and the enterprise form’s company-size buckets (1-20, 21-100, 101-500, 501-2k, 2k+) show Tavily sells from solo developers up through large enterprises. In February 2026, AI-cloud company Nebius agreed to acquire Tavily for a reported $275 million (up to $400 million on milestones).

Competitively, Tavily sits in the “web data and search for AI agents” category alongside tools such as Firecrawl, Exa, and other agent-retrieval APIs. Its differentiation is a simple, single-unit credit-based pricing model where each endpoint and search depth has a transparent per-request credit cost, so developers can forecast spend by counting requests rather than reasoning about compute or proxy fees.

Pricing summary : How Tavily’s credit-based search pricing works

Tavily uses a single credit-based model: every plan buys a monthly pool of API credits, each request burns credits according to its endpoint and search depth, and the per-credit rate falls as you move up the plan ladder. There are no seats — pricing is driven entirely by usage, with a free tier, pay-as-you-go, fixed monthly plans, and custom enterprise.

• Free tier (the on-ramp): Researcher gives 1,000 API credits/month with no credit card required.
• Pay-as-you-go (the overflow): $0.008 per credit, charged once your plan’s credit limit is reached; cancel anytime.
• Monthly plans (the commitment): Project $30 (4,000 credits, $0.0075/credit), Bootstrap $100 (15,000, $0.0067), Startup $220 (38,000, $0.0058), and Growth $500 (100,000, $0.005). The pricing page exposes these via a “Slide to adjust plan” selector on the Project card.
• Credits (the usage unit): Search costs 1 credit (basic) or 2 credits (advanced); Extract costs 1–2 credits per 5 successful URLs; Map costs 1–2 credits per 10 pages; Crawl is mapping + extraction; Research consumes a dynamic 4–250 credits per request depending on model.
• Enterprise (the quote): custom API calls, custom rate limits, enterprise-grade support/SLAs and security — priced by contact sales.

What makes this different: Tavily publishes a per-endpoint credit cost table and a volume-discount ladder (per-credit price drops from $0.0075 to $0.005), so a developer can estimate a bill from request counts while the vendor still captures volume upside through the named monthly plans and pay-as-you-go overflow.

Pricing by product

Tavily is a single web-search API billed in credits. The pricing page shows four headline cards (Researcher, Pay As You Go, Project, Enterprise) with a slider on the Project card that steps through the named monthly plans; the Credits & Pricing docs expose the full plan ladder and the per-endpoint credit costs. The tables below read directly from the live USD pricing page and the docs.

Tavily Search API (plan ladder)

Per-endpoint credit costs

Sales motions across products: PLG / self-serve for Researcher, Pay As You Go, and the Project–Growth monthly plans (sign up and subscribe online); sales-led for Enterprise (custom API calls, rate limits, security, and SLA).

Hidden costs : How Research requests and advanced depth quietly inflate the credit bill

Tavily’s headline rates look cheap — a basic search is a single credit — but the credit meter accelerates fast once an agent reaches for advanced search depth, multi-URL extraction, or the Research endpoint. The two archetypes below model realistic monthly spend at the pay-as-you-go rate of $0.008/credit.

A small team running a RAG chatbot that fires 100,000 searches a month, half of them advanced depth, plus light extraction:

At this volume the team is well past the Growth plan’s 100,000 credits, so the better path is a custom/enterprise quote — but the example shows how advanced-depth searches alone double the search line. Now a deep-research agent that runs 5,000 Research requests a month with model=pro:

Because a single model=pro Research call can consume up to 250 credits (~$2.00), a few thousand deep-research requests dwarf everything else on the bill — the dynamic 4–250 credit range is the most important number to budget for. The fix is to monitor the Usage endpoint, cap Research depth where possible, and move to a monthly plan or enterprise rate so the effective per-credit price drops toward $0.005.

Want to estimate your own Tavily bill? Use the Tavily pricing calculator to model your monthly cost based on search volume, search depth, and Research request mix.

Pricing evolution : From an open-source researcher to a $275M agentic-search acquisition

Tavily’s public credit-based pricing has been notably stable across every archived snapshot of tavily.com/pricing (2026-01 through 2026-05 are identical on price). The company’s bigger inflection points have been on the funding and ownership side rather than the rate card. The /pricing page was not archived before January 2026, so pre-2026 per-credit rates are recorded as unknown rather than guessed.

Cadence

Tracked range: 2023–2026. The /pricing page entered the Wayback archive only in 2026-01; the credit-based model itself dates to the 2024 commercial launch. Quarters not listed were verified stable (0 captured price changes, 0 SKU additions) within the available evidence.

Notable changes

• 2023 — Tavily originates as the open-source GPT Researcher project (no commercial pricing). (TechCrunch, Insight Partners.)
• 2024-07 — ~$5M seed round; commercial credit-based Search API goes live with a free monthly allotment and pay-as-you-go overflow. (TechCrunch funding coverage.)
• 2025-08 — $20M Series A led by Insight Partners and Alpha Wave Global; total funding reaches $25M; the enterprise page surfaces a ”🎉 Tavily raises $25M” banner (visible in the 2025-09 Wayback snapshot).
• 2026-01-23 — Earliest archived /pricing snapshot confirms Researcher (Free / 1,000), Pay As You Go ($0.008/credit), Project ($30 / 4,000) and Enterprise (Custom), with the docs ladder running to Growth ($500 / 100,000).
• 2026-02-10 — Nebius announces an agreement to acquire Tavily; Bloomberg reports a reported $275M, up to $400M on milestones. (Nebius newsroom; Bloomberg.)

The Nebius acquisition in detail

On 2026-02-10 Nebius announced an agreement to acquire Tavily to add real-time agentic search to its AI cloud platform, pairing Tavily’s web-access layer with Nebius’s Token Factory inference. Bloomberg reported the deal at a reported $275 million, with the figure potentially rising to $400 million if certain milestones are met. Founder and CEO Rotem Weiss is expected to join Nebius and continue leading the product. For pricing-watchers the key point is what did not change: through the captured window the public credit-based plan ladder ($0.008 PAYG, $30–$500 monthly, free 1,000-credit tier) stayed identical. The deal is a distribution and infrastructure bet — folding Tavily into a vendor that also sells inference — rather than a repricing event, at least at the time of writing.

What’s unique : One credit, every endpoint, and a published per-request price list

A single credit is the only unit you ever reason about. Unlike compute-priced crawlers that bill GB, CPU-seconds, or proxy bandwidth, Tavily collapses every operation — search, extract, map, crawl, research — into one credit-based meter. A developer estimates spend by counting requests and multiplying by a published credit cost, which is exactly the kind of forecastability that makes usage-based pricing easy to adopt.

The per-endpoint price list is public and granular. Tavily publishes the credit cost of every endpoint and search depth (basic search = 1, advanced = 2, extract = 1–2 per 5 URLs, and so on). That transparency is rare among agent-infra vendors and lets buyers model a bill before writing a line of code — the same playbook strong API metering businesses use to win developer trust.

You only pay for successful work. Failed URL extractions cost zero credits, and Extract and Map bill per 5 or 10 successful results. This “deliver-or-it’s-free” rounding aligns the meter with delivered value rather than attempts — a subtle but trust-building design choice for a product whose underlying web fetches sometimes fail.

Dynamic pricing on the Research endpoint exposes real cost variance. Rather than flat-rate a deep-research call, Tavily charges a dynamic 4–250 credits depending on model=mini vs model=pro and query complexity. It honestly reflects that multi-step research has variable cost — but it also means a single call can swing 60×, which is the main thing buyers must budget for.

The volume ladder is a true discount, not a packaging trick. Moving from Project ($0.0075/credit) to Growth ($0.005/credit) lowers the unit price by a third, and pay-as-you-go cleanly handles overflow at $0.008. The structure rewards commitment without locking buyers into annual contracts — closer to a pure-usage model than a seat-based one.

Strengths & weaknesses

Billing UX : Plan slider, monthly credit resets, and rate-limit handling

• “Slide to adjust plan” selector — the Project card on the pricing page carries a slider that steps the monthly credit allotment (and price) through the named plans: Project ($30 / 4,000 credits) → Bootstrap ($100 / 15,000) → Startup ($220 / 38,000) → Growth ($500 / 100,000).
• Monthly credit reset — API credits are allotted per month and reset on a monthly cadence (the pricing FAQ answers “When do my monthly API credits reset?”).
• Pay-as-you-go overflow — once a plan’s credit limit is reached, usage is billed at $0.008 per credit; PAYGO can be enabled and cancelled anytime, and production API keys require either an active paid plan or PAYGO enabled.
• Development vs Production API keys — rate limits are set by key environment: 100 RPM (development) vs 1,000 RPM (production) for the default endpoints.
• Per-endpoint rate limits — Crawl is capped at 100 RPM, Research at 20 RPM, and the Usage endpoint at 10 requests per 10 minutes, independent of the default limits.
• HTTP 429 + retry-after handling — exceeding a rate limit returns 429 Too Many Requests with a retry-after header (e.g. 60 seconds); Tavily recommends retry logic that respects it.
• Free for students program — a separate no-cost access path for students, applied for outside the standard plan flow.

Strategic wins : The pricing decisions that powered bottom-up developer adoption

1. Collapsing five endpoints into one credit unit

By metering Search, Extract, Map, Crawl, and Research all in the same credit, Tavily made spend trivially forecastable — a developer counts requests, not GB or CPU-seconds. This is the same legibility advantage that strong usage-based pricing metrics deliver, and it removes the single biggest source of bill anxiety for agent builders evaluating infra vendors.

2. A free 1,000-credit tier wired to open-source distribution

Tavily’s roots in GPT Researcher seeded a developer community before it ever charged a cent, and the free monthly allotment plus “Free for students” keeps the top of funnel wide. The result — over a million developers and 3M+ monthly SDK downloads — is exactly the product-led growth flywheel that made the company attractive enough for a $275M acquisition.

3. Publishing the full per-endpoint price list

Most agent-infra vendors hide unit economics behind a “contact us.” Tavily publishes the credit cost of every endpoint and depth, letting buyers model a bill before integrating. That radical price transparency lowers evaluation friction and signals confidence in the rate card.

4. A volume ladder that rewards growth without locking buyers in

The Project-to-Growth ladder cuts the per-credit price by a third while pay-as-you-go cleanly absorbs overflow at $0.008. Buyers get a genuine discount for scaling without signing an annual contract, which keeps the self-serve motion frictionless while still steering heavy users toward higher-revenue plans.

Areas to improve : Where Tavily’s credit model leaves money and clarity on the table

1. Surface Research-request cost variance on the pricing page

The 4–250 credit range on Research is the single biggest budgeting risk, yet it lives only in the docs. A worked example on the pricing page (“a typical model=pro call ≈ 120 credits ≈ $0.96”) would set expectations and reduce bill-shock support tickets — the kind of transparent communication that protects trust as usage scales.

2. Add a committed-use / annual discount tier

The ladder bottoms out at $0.005/credit on a monthly plan, with anything cheaper hidden behind enterprise sales. A published annual-commit discount (e.g. prepay 1M credits for a lower rate) would give high-volume teams a self-serve path to better economics and smooth Tavily’s own revenue — a common next step for maturing usage-based pricing models.

3. Unify the pricing page and the credit-cost docs

Plan prices live on /pricing while per-endpoint credit costs live in /docs, forcing buyers to cross-reference two surfaces to compute a bill. Folding a compact credit-cost table into the pricing page would close the gap and make the pricing page self-sufficient for evaluation.

Key takeaways

1. One meter beats many. Tavily collapses five distinct endpoints into a single credit, proving that a unified usage unit can make a multi-product API as forecastable as a single-product one. Buyers reason about request counts, not infrastructure primitives.
2. Transparency is a growth lever, not a giveaway. Publishing the full per-endpoint price list lowers evaluation friction and builds trust, helping Tavily reach 1M+ developers without a heavy sales motion.
3. Free tiers tied to open source compound. Seeding adoption through GPT Researcher and a generous free credit pool produced the developer base that made Tavily a $275M acquisition target.
4. Dynamic pricing is honest but needs guardrails. The 4–250 credit Research range correctly reflects variable cost, but without on-page worked examples it becomes the main source of buyer anxiety and bill-shock.
5. Stability is itself a feature. Across every archived snapshot Tavily never surprised buyers with a hike — predictable pricing is an underrated retention tool for usage-based products.

UBP implications

1. A single composite credit can unify a multi-endpoint API. Tavily shows that mapping every operation to one fungible credit — with published per-operation costs — gives buyers seat-like predictability inside a pure-usage model, a pattern other agent-infra vendors can copy.
2. Bill for successful results, not attempts. Charging zero for failed extractions and rounding per 5/10 successes aligns the meter with delivered value — a fairness signal that strengthens usage-based pricing’s credibility with developers.
3. Volume ladders can substitute for annual lock-in. A self-serve discount curve (here, $0.0075 → $0.005) rewards growth without contracts, suggesting usage-based vendors can defer sales-led commitments while still capturing volume upside.

Sources

• Tavily pricing page (accessed 2026-06-03)
• Tavily Credits & Pricing docs (accessed 2026-06-03)
• Tavily rate-limits docs (accessed 2026-06-03)
• Tavily enterprise / Talk to Sales page (accessed 2026-06-03)
• Tavily blog (accessed 2026-06-03)
• For the broader corpus of usage-based pricing teardowns, see the UsagePricing blueprint index.

Bottom line

Tavily turned an open-source research project into one of the cleanest usage-based pricing stories in agent infrastructure: a single credit spans every endpoint, the per-request price list is public, and you pay only for successful results — a model legible enough to recruit a million developers and predictable enough to survive a $275M acquisition without a repricing. The one number to watch is the Research endpoint’s 4–250 credit swing, the place where a tidy credit meter can still run away from you.

See how Tavily compares to other agent-infra and web-data vendors in the UsagePricing blueprint corpus.

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 snapshot — credit-based plan ladder unchanged

Jun 2026

Tavily prices its web-search API in credits: a free Researcher tier (1,000 credits/month), a $0.008/credit pay-as-you-go option, and monthly plans from Project ($30 / 4,000 credits) up through Bootstrap ($100 / 15,000), Startup ($220 / 38,000), and Growth ($500 / 100,000 credits) — with per-credit rates falling from $0.0075 to $0.005 as volume rises. Enterprise is custom/contact-sales. Footer now reads 'Trusted by 2M+ AI builders.'

Nebius agrees to acquire Tavily for a reported $275M (up to $400M)

Feb 2026

Nebius Group announced an agreement to acquire Tavily to add agentic search to its AI cloud platform; Bloomberg reported the deal at $275M, potentially rising to $400M on milestones, with founder Rotem Weiss joining Nebius. Public credit-based pricing was unchanged through the captured window (2026-01 to 2026-05 /pricing snapshots are identical on price). (Source: Nebius newsroom 2026-02-10; Bloomberg.)

First archived /pricing snapshot — four-card credit ladder

Jan 2026

The earliest Wayback capture of tavily.com/pricing (2026-01-23) shows the current structure: Researcher (Free / 1,000 credits), Pay As You Go ($0.008 / credit), Project ($30 / 4,000 credits, with a 'Slide to adjust plan' selector), and Enterprise (Custom). The docs ladder runs Project $30 → Bootstrap $100 → Startup $220 → Growth $500 with per-credit rates from $0.0075 down to $0.005.

$20M Series A led by Insight Partners (total funding $25M)

Aug 2025

Tavily raised a $20M Series A led by Insight Partners and Alpha Wave Global, bringing total funding to $25M. The enterprise page added a '🎉 Tavily raises $25M to power the Internet of Agents' banner (visible in the 2025-09 Wayback snapshot). No change to the public credit-based plan structure was captured. (Source: TechCrunch, Insight Partners; corroborated by the 2025-09 enterprise snapshot.)

Seed round + commercial credit-based API

Jul 2024

Tavily incorporated and raised a ~$5M seed round (mid-2024), launching the commercial Search API on a credit-based model with a free monthly allotment and pay-as-you-go overflow. Exact launch-day per-credit rates were not preserved in Wayback (the /pricing page was not archived until 2026), so pre-2026 rates are unknown. (Source: TechCrunch funding coverage.)

Origins — GPT Researcher open-source project

Jan 1970

Before any pricing existed, Tavily started as GPT Researcher, an open-source project Rotem Weiss built in 2023 to fetch real-time web data into LLM context windows. No commercial pricing yet; the product and its credit model came later. (Source: TechCrunch, Insight Partners.)