Braintrust pricing

About

Braintrust (braintrust.dev, legally Braintrust Data, Inc.) is an LLM evaluation, experimentation and observability platform. Engineering and product teams use it to log LLM traces, run systematic “evals” (scored experiments) against datasets, iterate on prompts in a playground, and monitor quality in production. The pitch is that disciplined evals turn a flaky AI feature into a shippable one — Braintrust cites teams moving accuracy from below 40% to over 80% in a few weeks.

The company was founded by Ankur Goyal (previously founder of ML-data startup Impira, and an early engineer/VP at MemSQL, now SingleStore). In October 2024 it raised a $36M Series A led by a16z’s Martin Casado (total funding ~$45M, reported ~$150M valuation), with operator-angels including Greg Brockman (OpenAI), Guillermo Rauch (Vercel), Simon Last (Notion), Arthur Mensch (Mistral) and Bryan Helmig (Zapier), plus Datadog and Databricks Ventures. Braintrust is used inside OpenAI, Notion, Stripe, Vercel, Airtable and Instacart.

Note: this is the AI evals company, not the unrelated “Braintrust” freelance talent network (usebraintrust.com / BTRST token).

Pricing summary : How Braintrust’s pricing model works

Braintrust charges a flat monthly platform fee per workspace — $0 (Starter), $249 (Pro), or custom (Enterprise) — that bundles a generous allotment of usage, then bills metered overages on four independent meters once you exceed the bundle:

1. Token credits for the AI-powered Topics/Loop pipeline: $10 included on Starter, $249 on Pro, then $0.06/mtok input and $0.40/mtok output on every tier. The Pro card shows the credit struck through from $100 to $249 — but the docs disclose that the $249 figure is a launch promotion until September 1, 2026, then $100 / month, so $100 is the standing allowance and $249 is the temporary one. Credits do not roll over month to month.
2. Processed data (total bytes ingested across logs, experiments and datasets): 1 GB / 5 GB included, then +$4/GB (Starter) or +$3/GB (Pro).
3. Scores (each recorded evaluation result counts as one): 10K / 50K included, then $2.50 per 1,000 (Starter) or $1.50 per 1,000 (Pro).
4. Retention storage — new on the pricing page as of July 2026. Starter’s 14-day window is fixed; Pro includes 30-day retention and then bills $0.50 per GB per month for data kept beyond that base window, up to a 180-day ceiling. Enterprise retention is custom, set with the account team up to 365 days. Experiments retain up to 365 days on every plan regardless of tier.

This is a textbook hybrid subscription-plus-usage model: a fixed platform fee that bundles usage, with metered overage beyond it.

What makes this different: unlike most LLM-observability rivals, Braintrust charges nothing per seat — every tier, including the free one, has unlimited users, projects, playgrounds, experiments and datasets. Cost scales with how much AI traffic and evaluation you run, not with team size. The flat $249 Pro fee (rather than a usage-only ramp) gives small teams a predictable ceiling before any overage kicks in. All four meters are billed on ingestion, not on what is currently stored, so deleting data or shortening a retention window does not reduce the month’s usage.

Pricing by product

Braintrust platform (self-serve plans)

*Pro’s $249 included Topics credit is a launch promotion until September 1, 2026, then $100 / month (Braintrust docs, Plans and limits). Topics credits do not roll over month to month, and the $0.06 / $0.40 per-mtok overage rates are uniform across plans.

Retention, entitlements and governance

Retention above covers logs and playground data. Experiments retain up to 365 days on all plans, including Starter. Starter’s 14-day window is documented as “not extendable”; Pro extends in 30-day increments up to the 180-day ceiling from Settings → Billing.

Basis caveat: Braintrust’s two docs pages describe the same $0.50/GB/mo charge on different bases — the billing FAQ says it “bills the data you ingest into each month you keep beyond the base window, not what’s currently stored,” while the retention docs say it “is billed on the highest amount retained during the billing month, so extending mid-month isn’t prorated.” The rate is identical on both surfaces; the measurement basis is not. Model conservatively until the vendor reconciles them.

Sales motions across products: self-serve / PLG for Starter and Pro (sign up, no credit card, upgrade in-product), sales-led for Enterprise (custom quote, security review, deployment choice).

Startup discount: qualifying early-stage startups — new Braintrust customers at Series A or earlier that have raised at least $100K in funding — can apply for 6–12 months of free Pro access. Legacy plans: accounts created before March 16, 2026 are grandfathered onto the feature set available at signup (custom and built-in permission groups, data-export automations, custom charts, environments, human review scorers), retained on upgrade but permanently lost on downgrade. Time-boxed model offer: through July 31, 2026, GLM-5.2 is a built-in model under the Braintrust provider and draws down the same monthly Topics credit; on Starter it requires on-demand usage to be enabled first. Vercel Marketplace: installs through Vercel are billed by Vercel, with the same usage limits and features as direct billing.

Hidden costs : What Braintrust users actually pay

The headline $249 looks cheap, but the three overage meters are where bills grow. The biggest surprise factor is processed data: it counts every byte of inputs, outputs, prompts, metadata, traces, spans and attachments — so verbose multi-step agents and large RAG contexts burn the GB allowance fast. Scores scale with how aggressively you sample production traffic for evals (the on-site calculator defaults to scoring ~15% of traffic). Token credits cover the platform’s own AI features (Topics/Loop), separate from your model bills.

A rough Pro-tier month for an AI-native team running real eval coverage (illustrative estimates, not official):

On-demand overage must be enabled; otherwise Starter simply pauses the Topics pipeline at the credit limit and caps usage rather than charging you. Enterprise pricing (custom retention, S3 export, on-prem, BAA/SLA) is unpublished and quote-only.

Want to estimate your own Braintrust bill? Use the Braintrust pricing calculator to model your costs based on usage patterns.

Pricing evolution : Braintrust pricing history and changes

Cadence

Tracked range: Jan 2025–2026 Q3. Quarters not listed above were verified stable (0 price changes, 0 SKU additions). The flat $249 Pro fee has held since launch; what has actually changed each time is the set of dimensions — seats removed, token credits added, retention priced.

Notable changes

• 2025 Q1 — Earliest model: free Builder (1,000 spans/week, ≤5 users) + custom Enterprise + free Open-source/.edu tier; self-serve pricing advertised as “coming soon.”
• ~Apr 2025 — Self-serve Free + Pro ($249/mo) launch. Switched the value metric to processed data (GB) and scores, with $3/GB and $1.50/1K overages on Pro. Free and Pro still capped at 5 users.
• ~Aug 2025 — Seat caps dropped to unlimited on Free and Pro; Pro gains unlimited trace spans. Braintrust commits to monetizing volume, not headcount.
• ~Mar 2026 — “Starter” rebrand + token credits. Free renamed Starter; a token-credit meter ($10 / $249 included, then $0.06/$0.40 per Mtok) added for the AI-powered Topics/Loop pipeline; transparent on-demand overage published.
• 2026-07-14 — Tier fees and all three live meters re-verified unchanged, but two previously unpublished commercial terms surfaced on the pricing page and billing FAQ: 6–12 months of free Pro for new customers at Series A or earlier that have raised at least $100K, and legacy-plan grandfathering for accounts created before 2026-03-16 (signup feature set retained on upgrade, permanently lost on downgrade).
• 2026-07-22 — Retention becomes the fourth meter. The comparison row was renamed from “Default retention” to “Included data retention”, and Pro’s cell now reads “30-day retention, then $0.50 per GB per month” — a rate the billing FAQ had referred to as a separate Pro charge without ever naming. Starter’s 14-day window stays fixed and Enterprise retention stays custom; every other rate held ($0 / $249 / custom fees, $0.06 and $0.40 per Mtok, $4 and $3 per GB, $2.50 and $1.50 per 1,000 scores). Like Braintrust’s other meters, retention storage is measured at ingestion, so deleting data or shortening a policy mid-cycle does not reduce that month’s charge.
• Scheduled: 2026-09-01 — Pro’s included Topics credit is set to fall from $249 to $100 per month. The pricing page presents $249 as a struck-through upgrade from $100, but the docs rate card labels it a “launch promotion until September 1, 2026, then $100 / month” — so the effective bundle shrinks by $149 of token credit on that date unless Braintrust extends it. The $249 platform fee and the $0.06 / $0.40 per-Mtok overage rates are unaffected.

What’s unique : Braintrust’s distinctive pricing mechanics

1. Zero per-seat pricing — unlimited users on every tier. This is the headline differentiator in a category (LangSmith, Arize, Humanloop) where most rivals charge per seat. Braintrust deliberately removed its early 5-user caps in mid-2025, betting that frictionless team sprawl drives more logged data and evals, which is what it actually bills for.

2. Four independent usage meters, not one. Cost is a function of token credits (its own AI features), processed data (GB ingested), scores (eval results) and, since 2026-07-22, retention storage on Pro — each with its own included bundle and its own overage rate that improves on Pro. Buyers tune cost by sampling rate, trace verbosity and how long they keep history, rather than by user count.

3. Every meter bills on ingestion, including the storage one. The $0.50/GB/mo retention charge is the clearest case: it is levied on data as it lands, so shortening a retention policy or deleting traces mid-cycle does not claw back that month’s cost. That makes retention behave like the other three meters rather than like conventional cloud storage, where buyers expect the bill to follow the current footprint — a distinction worth checking before modelling a Braintrust bill on per-GB storage pricing intuitions.

4. Flat fee as a predictability ceiling. The $249 Pro fee is a fixed, knowable number that bundles meaningful usage before any overage. Combined with spend alerts and a “pause-not-bill” default on Starter, it positions Braintrust against the bill-shock reputation of pure pay-as-you-go observability tools.

Strengths & weaknesses

Billing UX : Braintrust billing controls and transparency

• On-demand usage toggle — Overage billing is opt-in under Settings → Billing → On-demand usage. Without it, Starter pauses the Topics pipeline for the rest of the billing cycle at the credit limit rather than charging; with it enabled, excess usage is billed on the monthly on-demand invoice. No long-term commit is required.
• Spend alerts, but no hard cap — Braintrust states it does not offer hard spending limits that prevent usage. Pro and Starter can set up spend alerts at custom monthly-cost thresholds; Starter additionally gets automatic notifications at 80%, 90% and 100% of included usage limits, plus Topics-credit emails at 60% and 100% of the monthly credit.
• Retention extension control — Pro extends its window from Settings → Billing in 30-day increments up to the 180-day ceiling, at $0.50 per GB per month. Braintrust states extending mid-month is not prorated, so a mid-cycle bump is charged as a full month.
• Monitor usage dashboard — A Monitor usage view shows usage charts and detailed reports across the meters, including Topics token usage tracked separately from processed data and scores.
• Invoices and receipts — The Invoices dropdown under On-demand usage lists recent invoices and card validation fee receipts; Detailed usage breakdown opens the full invoice history in the billing portal.
• Tax preview at checkout — For taxable US billing addresses, sales tax is estimated at the Pro upgrade step: the upgrade preview shows an Estimated tax line that updates as you enter your billing address, collected with the prorated platform fee and itemized on the invoice.
• Dunning behavior — On a failed payment, Braintrust retries over a grace period with access uninterrupted and email notifications; if collection still fails, the account is downgraded to Starter and Pro can be restored by updating the card and resubscribing.
• Payment options — Self-serve Starter/Pro are credit-card, monthly; Vercel Marketplace installs are billed through the Vercel billing dashboard. Enterprise is invoiced under a custom contract (click-through DPA on Pro; custom DPA/BAA on Enterprise).

Strategic wins : Why Braintrust’s pricing decisions worked

1. Killing per-seat pricing to fuel land-and-expand

By removing seat caps in 2025, Braintrust let entire eng+product orgs onto the platform without a budget conversation — every new user generates more logs and evals, which is the metered revenue base. This is a textbook value-metric realignment: bill the thing that grows with value, not the thing that creates adoption friction. See how AI companies are shifting from per-user licenses.

2. A flat fee that tames bill-shock fear

A predictable $249 Pro fee with a generous bundle, plus opt-in overage and spend alerts, directly answers the cost-unpredictability and bill-shock anxiety that dogs pure usage-based observability tools. Buyers get a known floor and explicit controls before any variable spend.

3. Picking value metrics aligned to the eval workflow

Data ingested and scores recorded both rise with how seriously a team invests in evaluation — the exact behavior Braintrust wants to encourage. Tying price to evals run is a clean example of choosing the right usage metric.

4. Converting a feature gate into a priced meter

Until 2026-07-22, keeping eval history past 30 days meant an Enterprise contract — a binary that priced out mid-size teams whose only unmet need was history. Publishing $0.50/GB/mo on Pro replaces that cliff with a ramp: the buyer who needs 90 days of traces now pays tens of dollars instead of entering a quote cycle, and Braintrust captures revenue from a cohort that previously either churned or simply lost its data. It also converts retention from a support burden into a fourth expansion lever inside the existing self-serve tier — the same land-and-expand logic that removing seat caps served in 2025.

Areas to improve : Gaps in Braintrust’s pricing approach

1. Four meters are now harder to forecast, not easier

Token credits, processed data, scores and — since 2026-07-22 — retention storage each carry their own bundle and rate, and “processed data” counts every byte of traces. Adding a fourth dimension removed a packaging cliff but raised the modelling cost: a buyer must now project traffic volume, sampling rate, trace verbosity and history depth to land on a number. Publishing a per-trace or per-1,000-spans reference cost, or letting the on-site calculator output a single blended rate, would recover the legibility the extra meter spent. See bill shock and cost unpredictability.

2. Retention storage is metered at ingestion, which breaks buyer intuition

The $0.50/GB/mo rate is charged as data lands, not on what is actually being retained — so applying a stricter retention policy or deleting traces mid-cycle does not reduce that month’s bill. Buyers arriving from S3 or Datadog price it as data-at-rest and will model it wrong. Braintrust should either state the ingestion basis on the pricing page itself rather than only in the billing FAQ, or move retention onto a true stored-GB-month basis so the entitlement matches the mental model. Extending the same published rate to Starter would also give the free tier a way out of its fixed 14-day window.

3. Enterprise is a black box

On-prem, S3 export, BAA, SLAs and custom retention are all gated behind “contact us” with no published anchor. For a company selling transparency in AI, a more transparent Enterprise starting point would reinforce the brand.

Key takeaways

1. No per-seat fees is the defining choice. Unlimited users on every tier — including free — removes adoption friction and is rare among LLM-observability rivals that charge per seat.
2. Cost rides four meters, not one. Token credits, processed data, scores and retention storage each bill independently, so spend tracks AI traffic, eval intensity and history depth — never headcount.
3. The flat $249 Pro fee is a predictability anchor. It bundles real usage before overage and pairs with spend alerts and a pause-not-bill default to fight bill-shock.
4. The value metric evolved deliberately. Braintrust moved from spans and seats (2025) to processed data, scores and token credits (early 2026), then priced retention itself in July 2026 — each step binds the bill more tightly to evaluation volume and less to packaging tiers.
5. A feature gate is a pricing opportunity in disguise. Turning Pro’s 30-day retention ceiling into a $0.50/GB/mo meter monetized a need that previously forced a quote-only Enterprise jump, showing that the cheapest new revenue line is often an existing limit customers already want to cross.

UBP implications

1. Decouple adoption from monetization. Braintrust shows you can give away seats to maximize logged usage and still monetize cleanly on consumption — a strong pattern for developer-tool UBP.
2. A flat-fee floor de-risks consumption pricing. Bundling generous usage into a predictable subscription, then metering only the overage, is an effective way to sell usage-based pricing to budget-conscious buyers.
3. Unbundle your feature gates into meters before a competitor does. Braintrust’s 2026-07-22 retention change is the template: a hard limit that customers routinely wanted to exceed became a published per-GB rate, converting an Enterprise-only escalation into self-serve expansion revenue. The caveat for anyone copying it is basis — pricing storage on ingestion rather than on data at rest saves engineering effort but sells buyers a unit they will systematically mis-forecast in the LLM-observability space.

Sources

• Braintrust pricing page (accessed 2026-07-22)
• Braintrust pricing FAQ (accessed 2026-07-22)
• Braintrust billing FAQ — Topics credits, on-demand usage and retention storage (accessed 2026-07-22)
• Braintrust docs — Plans and limits (published rate card; discloses the Pro Topics-credit promotion end date and the 180-day Pro retention ceiling) (accessed 2026-07-22)
• Braintrust docs — Data retention (per-plan windows, 30-day extension increments, mid-month proration) (accessed 2026-07-22)
• Braintrust docs — Monitor usage (Topics-credit 60% / 100% emails, spend alerts, retention storage) (accessed 2026-07-22)
• Announcing our $36M Series A — Braintrust blog (Oct 8, 2024; accessed 2026-06-09)
• Internet Archive Wayback Machine snapshots of braintrust.dev/pricing, Jan 2025–Jun 2026 (accessed 2026-06-09)

Bottom line

Braintrust is the a16z-backed LLM evaluation and observability platform used inside OpenAI, Notion, Stripe and Vercel. Its pricing — free Starter, flat $249/mo Pro, custom Enterprise, with unlimited users on every tier and usage overages on token credits, processed data, scores and, since July 2026, retention storage at $0.50/GB/mo on Pro — is a deliberate land-and-expand play: give away seats, monetize evaluation volume. Pricing retention removed the biggest reason mid-size teams were forced into a quote cycle, but it also means the bill now rides four ingestion-based meters that need the on-site calculator to forecast, against a still-opaque Enterprise tier.

Want to compare Braintrust against other LLM-observability and MLOps companies? 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.

Retention storage becomes a priced meter on Pro ($0.50/GB/mo)

Jul 2026

The pricing page's retention row changed from an unpriced 'Default retention: 14 days / 30 days / Custom' to 'Included data retention', with Pro now published as '30-day retention, then $0.50 per GB per month'. Tier fees ($0 / $249 / custom) and the token-credit, processed-data and score rates were unchanged.

Re-verified; startup program + legacy-plan grandfathering surfaced

Jul 2026

Tier prices unchanged (Starter $0, Pro $249/mo, Enterprise custom; same three meters). Pricing page now advertises 6–12 months of free Pro for qualifying early-stage startups, and the billing FAQ documents legacy-plan grandfathering for pre-2026-03-16 signups plus GLM-5.2 drawing on the monthly Topics credit through Jul 31, 2026.

Pricing verified — credits + overage model

Jun 2026

Verified live: Starter $0, Pro $249/mo, Enterprise custom; token credits ($0.06/Mtok in, $0.40/Mtok out), processed-data ($4/$3 per GB) and score ($2.50/$1.50 per 1K) overages. Unlimited users on all tiers.

'Starter' rebrand + token credits + Topics/Loop

Mar 2026

Free was renamed 'Starter', and a token-credit meter was added for the Topics/Loop AI pipeline ($10 / $249 credits included, then $0.06/Mtok input, $0.40/Mtok output). Starter data overage set at $4/GB; scores $2.50/1K.

Seats go unlimited on Free and Pro

Aug 2025

The 5-user caps on Free and Pro were removed — every tier now advertises unlimited users, projects and experiments. Pro added unlimited trace spans; included quotas (1 GB / 5 GB data, 10K / 50K scores) held.

Self-serve Free + Pro ($249/mo) launches

Apr 2025

Braintrust shipped self-serve pricing: Free ($0, 1 GB data, 10K scores, 14-day retention, 5 users) and Pro ($249/mo, 5 GB then $3/GB, 50K scores then $1.50/1K, 5 users), alongside custom Enterprise.

Builder + Enterprise; self-serve 'coming soon'

Jan 2025

Earliest captured model: a free Builder tier (1,000 spans/week, up to 5 users) plus custom Enterprise and a free Open-source/.edu tier. The pricing page stated self-serve pricing was 'coming soon'.