Key Takeaways
- What Braintrust (AI evals) is — A platform for evaluating and monitoring LLM applications: experiments against datasets, built-in and LLM-as-judge scorers, a prompt playground, production tracing and an AI assistant that analyses traces. It stores traces in Brainstore, its own database. Not to be confused with the Braintrust talent marketplace.
- Where it stands in 2026 — An $80M Series B led by Iconiq at an $800M valuation in February 2026, with customers including Notion, Replit, Cloudflare, Ramp and Dropbox. Starter is free with 1 GB of processed data; Pro is $249 a month; keeping data in your own cloud (BYOC) needs Enterprise.
- The alternatives by job — Langfuse if you want open source and free self-hosting. LangSmith for LangChain and LangGraph stacks. Arize (Phoenix and AX) for ML and LLM monitoring in one place. Opik for Apache-licensed evals. Galileo for guardrail-style evaluation in production. Promptfoo for test-suite and red-team evals in CI, now owned by OpenAI.
- Bottom line — Braintrust has the most complete eval workflow in the category. Look elsewhere if you need self-hosting without an enterprise contract, framework-native tooling, or evals that live in CI rather than in a hosted platform.
What Braintrust (the AI eval platform) actually is
Braintrust is built around one loop: change a prompt or model, run it against a dataset, score the outputs, compare against the last run, ship, then pull failing production traces back into the dataset. Experiments, scorers (built-in, LLM-as-judge and code), a prompt playground and production tracing all serve that loop. Traces land in Brainstore, a database Braintrust built for querying large AI traces, and an in-product assistant analyses traces to suggest prompts and datasets.
In February 2026 Braintrust raised an $80M Series B led by Iconiq at an $800M valuation, naming Notion, Replit, Cloudflare, Ramp and Dropbox as customers.
When an alternative is the better fit
- You need traces in your own infrastructure without an enterprise contract. Braintrust's BYOC is Enterprise-only; Langfuse, Opik and Phoenix self-host for free.
- Your app is built on LangChain or LangGraph. LangSmith instruments it with almost no work.
- You also monitor classic ML models. Arize covers both in one platform.
- You want evals to gate merges in CI. Promptfoo runs evals and red-team tests as config in your pipeline.
- You need runtime guardrails, not just offline evals. Galileo focuses on evaluation models that run in production.
The 6-tool shortlist
- Langfuse: open-source tracing, evals and prompt management; MIT-licensed core; part of ClickHouse since January 2026. Cloud from $0, Core $29 and Pro $199 a month.
- LangSmith: LangChain's tracing, evals and prompt hub; strongest for LangChain and LangGraph apps; self-hosting on Enterprise.
- Arize (Phoenix and AX): open-source Phoenix for tracing and evals on OpenTelemetry, commercial AX for enterprise ML and LLM observability.
- Opik (Comet): Apache-licensed tracing, evals and automated prompt optimisation; self-host or Comet cloud.
- Galileo: evaluation and guardrail platform built around its own evaluator models for production monitoring.
- Promptfoo: open-source CLI for evals and red-teaming in CI; acquired by OpenAI in March 2026.
Comparison matrix
| Feature | Braintrust | Alternatives shortlist |
|---|---|---|
| Evaluation workflow | ||
| Experiments and datasets | Core of the product: compare runs side by side, diff scores, promote production traces into datasets | Langfuse, LangSmith and Opik all have datasets and experiments; Promptfoo runs them as config-driven test suites |
| Scorers | Built-in scorers, LLM-as-judge and custom code scorers | All six support LLM-as-judge; Galileo leans on its own evaluation models; Promptfoo adds red-team graders |
| Prompt playground | Version-controlled prompts tested against production data | Langfuse, LangSmith and Opik have playgrounds; Promptfoo is CLI and config first |
| Deployment and ownership | ||
| Self-hosting | SaaS on all plans; BYOC (data in your cloud) on Enterprise only | Langfuse, Opik, Phoenix and Promptfoo self-host freely; LangSmith self-host on Enterprise |
| Open source | Commercial platform; SDKs and autoevals library are open source | Langfuse (MIT core), Opik (Apache 2.0), Phoenix and Promptfoo are open source |
| Independence | Independent, venture-backed ($800M valuation, Feb 2026) | Langfuse is part of ClickHouse (Jan 2026); Promptfoo part of OpenAI (Mar 2026); LangSmith is LangChain's |
| Pricing | ||
| Free tier | Starter: 1 GB processed data, 10k scores a month, 14-day retention | All have free tiers or free self-hosting |
| Paid entry point | Pro $249/month: 5 GB data, 50k scores, 30-day retention; then $3/GB and $1.50 per 1k scores | Langfuse Core $29 and Pro $199 a month; LangSmith per seat; Galileo and Arize AX by quote at scale |
Decision framework
- Eval workflow is the priority and SaaS is fine? Stay on Braintrust.
- Need free self-hosting and open source? Langfuse or Opik.
- Built on LangChain or LangGraph? LangSmith.
- Monitoring classic ML and LLMs together? Arize.
- Evals must gate CI? Promptfoo.
- Need production guardrails? Galileo.
Frequently asked questions
What is Braintrust?
Braintrust is an AI evaluation and observability platform. Teams use it to run experiments that score LLM outputs against datasets, compare prompt and model changes side by side, trace production requests including tool calls and retrieved context, and turn production traces into new test cases. It stores traces in Brainstore, a database it built for AI trace queries. The company raised an $80M Series B at an $800M valuation in February 2026.
What are the best Braintrust alternatives?
Langfuse for an open-source platform you can self-host for free. LangSmith for teams on LangChain or LangGraph. Arize (open-source Phoenix plus the commercial AX platform) when you monitor both classic ML and LLMs. Opik from Comet for Apache-licensed evals and prompt optimisation. Galileo for evaluation models and guardrails in production. Promptfoo for evals and red-teaming as code in CI.
Is there an open-source alternative to Braintrust?
Yes. Langfuse (MIT-licensed core), Opik (Apache 2.0), Arize Phoenix and Promptfoo are all open source and self-hostable. Langfuse and Opik are the closest in shape to Braintrust, with datasets, experiments, scorers, prompt management and tracing in one UI. Promptfoo covers the CI test-suite side of evals rather than the hosted experiment workflow.
How much does Braintrust cost?
Braintrust Starter has no platform fee and includes 1 GB of processed data and 10,000 scores a month with 14-day retention, then $4 per GB and $2.50 per 1,000 scores. Pro is $249 a month with 5 GB, 50,000 scores and 30-day retention, then $3 per GB and $1.50 per 1,000 scores. Enterprise is custom, with retention up to 365 days and BYOC deployment.
Braintrust vs Langfuse?
Braintrust is eval-first and commercial; Langfuse is observability-first and open source. Braintrust has the more refined experiment and scoring workflow. Langfuse self-hosts for free and is cheaper to start, but is now part of ClickHouse. See our full Langfuse vs Braintrust comparison.
Braintrust vs LangSmith?
Both are commercial eval and tracing platforms. LangSmith is the obvious choice if your application is built on LangChain or LangGraph, because tracing is nearly automatic there. Braintrust is framework-agnostic and stronger on experiments and scoring. LangSmith offers self-hosting on Enterprise; Braintrust offers BYOC on Enterprise.
What happened to Promptfoo?
OpenAI announced it was acquiring Promptfoo in March 2026 to strengthen security testing for AI agents. Promptfoo remains an open-source CLI for evals and red-teaming. If independence from model vendors matters for your eval stack, that ownership change is worth weighing.
Ready to Find the Right AI Tools?
Browse our data-driven rankings to find the best AI tools for your team.