The best Langfuse alternatives are Arize Phoenix if you want another free self-hosted tracer, LangSmith for LangChain and LangGraph teams, Braintrust for a polished hosted eval workflow, and W&B Weave if you also want training and serving on one cloud; Kojen is an alternative only for the evaluation half, adding confidence intervals, pass^k, red-teaming and tuning, not production tracing. Langfuse itself remains MIT-licensed and free to self-host after ClickHouse announced it was acquiring the company on January 16, 2026.
Most teams leave Langfuse because they want per-item repetitions, statistics or automatic improvement, which it doesn't offer, rather than because of the acquisition.
Phoenix is free to self-host with no feature gates and is OpenTelemetry-native. It supports experiment repetitions, which Langfuse doesn't. Its Elastic License 2.0 is source-available rather than OSI open source.
LangSmith traces any framework but understands LangChain deeply, adds annotation queues, repetitions with standard deviations, prompt optimization jobs and hosting for agents. It's priced per seat, and self-hosting is an Enterprise add-on.
Pricing: Developer $0 (1 seat, 5k base traces/month); Plus $39/seat/month (10k traces); Enterprise custom. Usage beyond that is pay as you go. Source · Open source: no
Strengths
The natural choice for LangChain and LangGraph teams, with production hosting for agents.
Strong tracing, annotation queues and prompt management.
Cloud, hybrid and self-hosted deployment options.
Limits
Per-seat pricing, and the free tier has one seat.
Reliability stats stop at mean and standard deviation; no intervals or pass^k.
Best for: a polished hosted eval and experiment workflow
Braintrust's experiments, scorers and Loop agent are among the best-built, with unlimited users on the free tier and a gateway. Self-hosting is Enterprise-only.
Pricing: Starter $0 (unlimited users, 1 GB, 10k scores, 14-day retention); Pro $249/month; Enterprise custom. Source · Open source: no
Strengths
A mature, polished eval and tracing product, with unlimited users even on the free tier.
Loop works across logs, experiments, datasets and scorers in one place.
Gateway, environments and online scoring of production traffic in the same product.
Limits
No confidence intervals or significance tests, and no pass@k beyond grouping trials.
Measures and suggests, but doesn't train weights, red-team or serve models.
Best for: evaluating, training and serving on one cloud
W&B Weave, now inside CoreWeave Forge, pairs tracing and evals with serverless RL and SFT into LoRA adapters and OpenAI-compatible serving. Weave data allowances are small.
Pricing: Free (up to 5 seats, 1 GB/month of Weave data); Pro from $60/month for companies under 50 people; Enterprise custom. Source · Open source: yes (Apache-2.0 (SDK))
Strengths
Covers evaluate, train (RL and SFT to LoRA) and serve on one cloud.
Apache-2.0 SDK and W&B's mature experiment tracking.
Guardrails that double as production monitors.
Limits
Small Weave data allowances with per-MB overage; Pro is limited to companies under 50 people.
Branding and URLs are in flux after the move into CoreWeave Forge.
Best for: evaluations you can trust, plus improvement
Kojen is not a tracing tool: it has no SDK for apps running elsewhere. It replaces the evaluation side of Langfuse with confidence intervals and pass^k from several samples per case, adds red-team simulations, and tunes and deploys the model. We make Kojen.
Pricing: Free to start, no card needed; then pay as you go. GPU training (LoRA) is paid in credits billed by the second at the GPU's live price. Team and volume pricing by email. Source · Open source: no
Strengths
Measures and improves in one loop: evaluations, red-teaming, prompt tuning, model search and deployment share the same project and test cases.
Honest statistics by default: every rate has a confidence interval, tuned prompts must beat the current one in a paired test, and pass@k / pass^k come from several samples per case.
Agent-native: the full loop is exposed over an MCP server and a REST API with an OpenAPI spec, with the same 24 tools as the in-app agent.
Limits
Young product: weight training is early access on 8B models, LoRA adapters can't be served yet, and full fine-tuning and RLHF are previews.
No tracing SDK for apps running elsewhere, so it is not a replacement for an observability platform.
Closed source and hosted only; no self-hosting.
Side-by-side comparison
"Partial" means the feature exists with real limits, given in the note. "Not verified" means we couldn't confirm it from the vendor's own material.
Langfuse alternatives: features and pricing, checked November 2026
Checked Nov 15, 2026
Langfuse
Arize Phoenix
LangSmith
Braintrust
W&B Weave
Kojen
What it is
Evaluation & observability
ClickHouse (acquisition announced January 16, 2026); still MIT open source
Evaluation & observability
Arize AI (independent)
Evaluation & observability
LangChain (independent)
Evaluation & observability
Braintrust Data (independent)
Evaluation & observability
Weights & Biases, part of CoreWeave since May 2025; sold inside CoreWeave Forge since September 2026
Training dojo
Kojen (that's us)
Pricing
Cloud: Hobby free (50k units/month); Core $29/month; Pro $199/month; Enterprise $2,499/month. Self-hosted open source: free. Source
Free (up to 5 seats, 1 GB/month of Weave data); Pro from $60/month for companies under 50 people; Enterprise custom. Source
Free to start, no card needed; then pay as you go. GPU training (LoRA) is paid in credits billed by the second at the GPU's live price. Team and volume pricing by email. Source
Run a model or app over a set of test cases and score the results.
Yes
Experiments on datasets
Yes
Yes
evaluate() on datasets
Yes
Eval() experiments on datasets
Yes
weave.Evaluation
Yes
Test cases scored for accuracy, hallucination, bias and latency
LLM-as-judge scoring
A model grades outputs against a rubric or reference.
Yes
Managed evaluator catalog and custom judges
Yes
phoenix-evals
Yes
LLM evaluators from the UI or SDK
Yes
autoevals library plus custom scorers
Yes
Scorers and guardrails
Yes
Judge kept out of the target's model family in simulations
pass@k / pass^k
Several samples per case, reported as any-success (pass@k) or all-success (pass^k).
No
Experiments assume no repetitions per item
Partial
Experiment repetitions with an average; no pass@k
Partial
num_repetitions with mean and standard deviation; no pass@k
Partial
trialCount reruns each input; no pass@k metric
Partial
trials per example; no pass@k
Yes
1–10 samples per case
Confidence intervals
Rates reported with an interval or a significance test, not just a point estimate.
No
Averages and deltas
No
Box plots and diffs in AX
No
Standard deviation only
No
Score deltas and improvement/regression counts
No
Aggregate charts and deltas
Yes
Wilson and bootstrap intervals on every rate
Prompt optimization
Automatically rewrites instructions or examples to raise a metric.
No
Prompt versioning and A/B labels, no optimizer
Partial
Prompt Learning in Arize AX, not Phoenix
Partial
Optimization jobs (promptim) via API; Polly edits prompts on request
Partial
Loop agent suggests prompt edits you accept
No
Logs DSPy optimizer runs; no optimizer of its own
Yes
Prompt tuning plus search across 11 hosted models
Red-teaming
Generates adversarial attacks (prompt injection, jailbreaks, data leaks) and scores them.
No
Partial
Documented workflow and PyRIT cookbook; no attack generator
No
No
Only a prompt-injection-detector cookbook
Not verified
Safety guardrails, no attack generator found
Yes
Prompt injection, jailbreak, data exfiltration, harmful content, PII leak
Production tracing
Logs and traces live traffic from your app.
Yes
OpenTelemetry-based tracing
Yes
OpenTelemetry-native
Yes
Yes
Tracing and online scoring
Yes
Partial
Request logs, latency, errors and feedback for Kojen endpoints only; no tracing SDK
Official MCP server
An MCP server from the vendor that lets agents drive the product.
Yes
Authenticated MCP server, read and write
Yes
Remote MCP endpoint in Phoenix
Yes
Python server and hosted remote MCP with OAuth
Yes
Hosted MCP server
Yes
wandb-mcp-server
Yes
ai.kojen/kojen in the MCP Registry, OAuth sign-in
Self-host
You can run it on your own infrastructure.
Yes
Free, no core-feature limits
Yes
Free with no feature gates
Partial
Enterprise add-on
Partial
Enterprise only; you run the data plane, Braintrust the control plane
Partial
Needs a licensed W&B Server
No
Hosted only
Langfuse after ClickHouse
ClickHouse announced it was acquiring Langfuse on January 16, 2026. Langfuse's FAQ says no licensing changes are planned, self-hosting stays a first-class path and Langfuse Cloud keeps running as is. As of November 2026 the cloud plans run from a free Hobby tier to $199 a month for Pro, and self-hosting the open-source edition is free with all core features.
Langfuse's own docs say experiments assume no repetitions per item, and comparisons show averages and deltas. If you need to know whether an agent succeeds every time (pass^k) or whether a change is real (an interval), that is the gap to fill.
When to choose Langfuse, and when to choose Kojen
Choose Langfuse if…
You want MIT-licensed, free, unrestricted self-hosting.
You need production tracing at low cost with long retention.
You want an MCP server that can write prompts, datasets and scores.
Choose Kojen if…
You need repeated samples per case, pass^k and confidence intervals.
You want red-team simulations next to your evaluations.
You want the tool to tune the prompt or pick the model, not just score it.
Frequently asked questions
Is Langfuse still open source?
Yes. Langfuse is MIT-licensed apart from its ee/ folders, and its FAQ after the ClickHouse acquisition says no licensing changes are planned.
What is the best free self-hosted alternative to Langfuse?
Arize Phoenix: free to self-host with no feature gates and OpenTelemetry-native, though under the source-available Elastic License 2.0 rather than an OSI license.
Does Langfuse support pass@k?
No. Its docs say experiments currently assume no repetitions per dataset item. Phoenix, LangSmith and Braintrust can repeat runs; Kojen and Inspect compute pass@k and pass^k.
Can Kojen replace Langfuse?
Only for evaluation. Kojen logs requests to its own deployments but has no tracing SDK for apps running elsewhere, so keep Langfuse or another tracer for production observability.
Run an LLM-judged evaluation with confidence intervals, a red-team pass and a prompt-tuning run on your agent. Free to start, no card needed, then pay as you go.