Eval-first platform for AI products: experiments, datasets, playgrounds and OTel tracing, free at 1 GB processed data and 10k scores a month.
Teams whose bottleneck is measurement rather than dashboards: Braintrust charges for processed data and scores instead of seats, so unlimited users, projects, datasets, playgrounds and experiments are on every plan including the free Starter tier. That makes it workable where product managers, domain reviewers and engineers all need to look at the same experiment. It suits anyone instrumenting with OpenTelemetry, since traces and logs can be sent over OTLP/HTTP and LLM calls are converted into Braintrust LLM spans that can be saved as prompts and evaluated directly.
The billing unit punishes verbose agents. You pay per gigabyte of processed data - $4/GB on Starter, $3/GB on Pro - so long contexts, large tool payloads and retrieved documents drive cost regardless of how few requests you make, and scores are metered separately at $2.50 per 1,000 on Starter and $1.50 per 1,000 on Pro. Retention is short by default: 14 days on Starter, 30 days on Pro, and stretching Pro to 180 days costs an extra $0.50/GB/month, with up to 365 days only on Enterprise. Self-hosting is Enterprise-only and is not a full air-gap - the data plane runs in your cloud, but the Braintrust-hosted control plane still provides the web UI, authentication, user management and project and experiment metadata, with auth credentials held at Clerk. SAML SSO, custom RBAC roles, a BAA, S3 data export, custom retention policies and uptime SLAs are all Enterprise-only, and Enterprise is quote-only.
Pros
Cons
Neither. Users are unlimited on every plan, and the metered units are processed data in gigabytes, scores per 1,000, and model credits in tokens. Starter includes 1 GB, 10K scores and $10 of credits; Pro includes 5 GB, 50K scores and $100 of credits. Overages run $4/GB and $2.50 per 1,000 scores on Starter, and $3/GB and $1.50 per 1,000 scores on Pro. Unused credits do not roll over.
Logs and playground data are retained 14 days on Starter and 30 days on Pro. Pro retention can be extended up to 180 days for an additional $0.50 per GB per month. Enterprise retention is custom, up to 365 days. Experiments are an exception and are retained up to 365 days on all plans, which matters if your evaluation history is the asset you care about keeping.
Only on Enterprise, and only partly. The data plane holds all sensitive data - experiments, logs, traces, spans, datasets, prompts and completions - and runs the Braintrust API, PostgreSQL, Redis, object storage and Brainstore in your AWS, GCP or Azure account. The web UI, authentication, user management and metadata such as project and experiment names stay in the Braintrust-hosted control plane, with credentials at Clerk.
Yes, for traces and logs over OTLP/HTTP, with JSON or protobuf payloads; gRPC is not documented. Traces go to api.braintrust.dev/otel/v1/traces in the US or api-eu.braintrust.dev/otel in the EU, authenticated with a Bearer API key plus an x-bt-parent header naming a project_id, project_name or experiment_id. Self-hosted logs need data plane v2.12.0 or later, and span-event ingestion needs v2.14.0.
Facts sourced from the vendor’s own site, last verified 2026-10-10.
The Weekly Source
One email a week. New listings, repriced tools, and the ones we would skip - including this one if its pricing changes.
