Langfuse vs Arize Phoenix in 2026 - License vs Eval Depth

Langfuse and Arize Phoenix are the two names that come up when someone asks for the open-source LLM observability default. Both are free to self-host. Both do tracing and evals. Both are framework-agnostic. So why do people pit them against each other? Because they are strong on opposite axes. Langfuse wins on license clarity and cheap managed scaling. Phoenix wins on RAG eval depth and OpenTelemetry-native architecture. Pick wrong and you either fight a license you did not read, or you give up the best RAG scoring in the category. I have added Opik as a third option, because it splits the difference on the one thing both of these leave on the table.

The short version

Tool	Best for	Self-host / license	OpenTelemetry
Langfuse	Cleanest license, cheap managed scaling	Free, MIT, near-complete	OTLP backend (HTTP only)
Arize Phoenix	Best RAG eval, fastest to try	Free, but ELv2 server	Native (OpenInference)
Opik	Permissive OSS + cheapest cloud	Free, Apache-2.0, full features	One ingestion path

Langfuse: the clean license and the cheap scale

Langfuse is the tool “what’s the self-hostable alternative” almost always resolves to, and the license is a big reason. It is MIT - genuinely OSI open source - and self-hosting is free with only three features enterprise-gated. Tracing, evals, prompt management, human annotation and RBAC are all in the free build. There is no source-available asterisk to read around.

The other half is cost at scale. At 1M events a month it runs about $101/mo managed, and self-hosted the per-trace cost disappears. Core is $29/mo for 100k billable units with 90-day retention. It runs as an OpenTelemetry backend mapping the GenAI semantic conventions.

The catch is operational. Langfuse v3 needs Postgres plus ClickHouse, Redis and S3-compatible storage - four services - and the migration is where self-hosters get stuck. Its OTLP support is also HTTP only, no gRPC yet. And it is now a ClickHouse subsidiary after the January 2026 acquisition, which I would weigh for a multi-year bet. The Langfuse page has the full detail.

Arize Phoenix: fastest to try, best RAG eval, read the license

Arize Phoenix is the OSS pick that starts fastest - a working trace UI on your laptop in under a minute. It is genuinely OpenTelemetry-native, built on OTel and Arize’s own OpenInference conventions, so it is framework-agnostic by design. Its 50+ pre-built eval metrics include what reviewers call the best RAG evaluation in the category - real retrieval and answer scoring with no judge prompts to write. If eval, and RAG eval specifically, is the reason you are here, this is where Phoenix pulls ahead of Langfuse.

The gotcha is the license, and it is a real one. Arize markets Phoenix as “fully open source, no feature gates,” but the main server repo is Elastic License 2.0 - source-available, not OSI-approved open source. Only the client and eval subpackages (phoenix-evals, phoenix-client, phoenix-otel) are Apache-2.0. ELv2 forbids offering Phoenix as a hosted or managed service to third parties. For internal use it behaves like open source and the features are not gated. But if your plan is to resell it as a service, read the license first. There are also reports of ingest lag before traces appear. One more note: the free product is Phoenix, but Arize AX is a separate managed platform, and its pricing page was unreadable at verification, so budget a sales call if you want the cloud. Arize the company raised a $70M Series C in February 2025, so it is not going anywhere.

Opik: the middle path both of these miss

Langfuse gives you a clean license but thinner RAG eval. Phoenix gives you deep eval but a source-available server. Opik is the one that gives you a permissive license and a cheap cloud together. Its OSS build is Apache-2.0 with the full feature set self-hosted - unlimited spans, members and retention, no gates - more permissive than Phoenix’s Elastic License. On the cloud, Pro is $19/mo for 100k spans, the cheapest paid cloud of the major platforms. Comet builds it, it is the fastest-growing project of its peers at roughly 20.8k GitHub stars, and it ships LLM-as-judge, code metrics, guardrails and an Agent Optimizer.

Its gotcha is per-seat pricing at scale, and OTel is one ingestion path here rather than the native architecture that Phoenix is built on. If strict OTel-native matters, Phoenix still wins that specific axis.

Langfuse vs Phoenix: which should you pick?

• You need OSI-approved open source specifically - Langfuse. MIT settles it. Phoenix’s ELv2 server is source-available, which is a different thing.
• RAG eval is the whole reason you are here - Arize Phoenix. The 50+ metrics and best-in-class RAG scoring are the differentiator.
• You want to try something in the next 60 seconds - Phoenix, which runs locally in under a minute.
• You want the cheapest managed cloud with a clean license - Opik at $19/mo, ahead of Langfuse Core at $29/mo.
• You plan to resell your observability stack as a service - Langfuse or Opik. Phoenix’s ELv2 rules that out.
• You want strict OpenTelemetry-native architecture - Phoenix, built on OTel and OpenInference from the ground up.

Honestly, most teams can be happy on either. If you value a clean license and cheap scaling, Langfuse. If you value eval depth and an OTel-native core, Phoenix. If you want a permissive license and the cheapest cloud in one tool, Opik is the pick that quietly beats both on those two axes. Every price and date here was read from each vendor’s own materials, verified on 26 July 2026, and this category ships breaking changes monthly, so we re-verify every 30 days.