Galileo vs Langfuse in 2026 - Enterprise Eval Intelligence or Open-Source Default?

First, the disambiguation, because it matters more here than for any other comparison. The Galileo in this post is the LLM eval platform at galileo.ai, not the Google-acquired text-to-UI design tool of the same name. If a review mentions “$39 a month” or “your designs are publicly visible,” that is the wrong company. Check the domain on every source.

With that cleared up - these two sit at opposite ends of the category. Galileo is the best-funded, most research-forward eval platform, built around proprietary Luna models and real-time guardrails, and sold through a rep. Langfuse is the open-source default you can self-host free and forecast to the dollar. One is enterprise eval intelligence. The other is open-source control. I have added Arize Phoenix as the third option, because if your real need is deep eval on open-source terms, it belongs in the conversation.

The short version

Tool	Best for	Starting price	Self-host	License
Galileo	Enterprise eval intelligence + guardrails	Free / $100/mo	Enterprise only	Closed
Langfuse	Open-source default, cheap at scale	Free / $29/mo	Free, near-complete	MIT
Arize Phoenix	Fast OSS tracing and best RAG eval	Free (OSS)	Free, runs locally	ELv2 server

Galileo: eval intelligence and real-time guardrails

Galileo calls its category “Evaluation Intelligence.” It captures traces, scores them with research-backed metrics, and turns those offline evals into real-time production guardrails. The technical bet is its proprietary Luna and Luna-2 eval models - small models fine-tuned for tasks like hallucination, context adherence and prompt-injection detection. The argument is economic - if your evaluator is cheap and fast enough, you run it on every request instead of sampling, which is what turns offline evals into live guardrails. Galileo claims Luna is up to 11x faster and 97% cheaper than a GPT-3.5-based judge, though those are vendor benchmarks - treat them as claims until you test on your own traffic. It is the best-funded platform in the space at roughly $68M raised, so it has the lowest bankruptcy risk, and the free tier is genuinely generous at 5,000 traces with unlimited users.

The reservations are commercial. Above the $100/mo Pro tier (50,000 traces, billed yearly) everything is contact-sales, and self-host - VPC or on-prem - is Enterprise-only with no open-source version. The recurring, credible complaint is that you cannot size ROI without talking to a rep. This is a platform you buy through a sales process, not a card.

Langfuse: the open-source default

Langfuse is the open-source default for LLM observability, and its whole pitch is self-host that does not cripple you. Only three features are enterprise-gated in the self-host build - tracing, evals, prompt management and RBAC are all free under MIT. It is framework-agnostic, runs as an OpenTelemetry backend on an OTLP endpoint, and the economics are the headline - at 1M events a month it runs about $101/mo managed, roughly 25x cheaper than LangSmith for comparable volume, and self-hosting removes the per-trace cost entirely. Pricing is transparent all the way up, no sales call required.

Its catch is operational, not commercial. Langfuse v3 needs Postgres plus ClickHouse, Redis and S3-compatible storage - four services - and the migration to that architecture is where self-hosters get stuck. And where Galileo hands you research-grade eval models and guardrails out of the box, Langfuse gives you the eval primitives and expects you to assemble more of the orchestration yourself. It is now a ClickHouse subsidiary after the January 2026 acquisition, worth filing away for a multi-year bet.

Arize Phoenix: open-source eval depth

If your priority is eval depth on open-source terms, Arize Phoenix belongs here. It ships 50+ pre-built eval metrics and the best RAG evaluation in the category, runs locally in under a minute, and is OpenTelemetry-native. Phoenix OSS is free with no usage caps.

Its catch is the license. The server repo is Elastic License 2.0 - source-available, not OSI open source - and ELv2 forbids offering Phoenix as a hosted service to third parties. For internal use it behaves like open source and the features are not gated. It is the eval-depth specialist between Galileo’s guardrails and Langfuse’s general-purpose openness.

Galileo vs Langfuse: which should you pick?

• You are an enterprise burned by production hallucinations and want research-grade eval plus real-time guardrails - Galileo, and confirm you have the right company before the sales call.
• You want transparent pricing you can commit to without a rep - Langfuse. Galileo goes sales-led above $100/mo.
• You need data on your own infrastructure without a contract - Langfuse, which self-hosts free. Galileo self-host is Enterprise-only.
• You want the cheapest platform at production trace volume - Langfuse, roughly 25x cheaper than LangSmith and self-hostable free.
• Your real need is deep eval, RAG especially, on open-source terms - Arize Phoenix, as long as you are not reselling it.

The honest case for Galileo: if continuous, guardrail-grade evaluation in production is the problem - and you have the budget and the patience for an enterprise motion - the Luna-plus-guardrails loop is a genuinely stronger story than anything Langfuse gives you off the shelf. The Luna speed and cost multipliers are vendor claims, but the direction is right. Langfuse wins on openness, price and control; Galileo wins on eval intelligence and staying power.

For the wider field, see our Galileo alternatives and Langfuse alternatives breakdowns, or the head-to-head Galileo vs Arize Phoenix. Every price and date here was read from each vendor’s own materials on 26 July 2026. This category ships breaking changes monthly, so we re-verify every 30 days.