5 Galileo Alternatives With Real Self-Host and Public Pricing (2026)

Galileo is the best-funded platform in this category, at roughly $68M raised, and the most research-forward. Its Luna and Luna-2 eval models are a genuine bet on making evaluation cheap enough to run on every production request. The free tier is real too - 5,000 traces a month with unlimited users, which is rare at $0.

That is not why people go looking for alternatives. The problem is that everything serious lives behind a sales call. Above the $100/mo Pro tier, pricing is contact-sales, and self-host - VPC or on-prem - is Enterprise-only with no open-source version at all. If keeping trace data on your own infrastructure is a hard requirement, you are in a sales process from the first day. The recurring, credible complaint about Galileo is exactly that you cannot get a real price or size ROI without a rep.

So this is not a “Galileo is bad” list. The Luna models and the guardrails loop are a real, differentiated bet, and if you are an enterprise that got burned by production hallucinations, it earns the look. This is a list for the people who want transparent pricing they can commit to on a card, or open-source self-hosting without a contract. I have matched each pick to the reason you are here.

The short version

Tool	Best for	Starting price	Self-host	License
Galileo	Enterprise eval intelligence and guardrails	Free / $100/mo	Enterprise only	Closed
Langfuse	Open-source default, cheap self-host	Free / $29/mo	Free, MIT, near-complete	MIT
Opik	Cheapest managed cloud, permissive OSS	Free / $19/mo	Free, full features, no gates	Apache-2.0
Arize Phoenix	Fast OSS tracing and the best RAG eval	Free	Free, but ELv2 server	ELv2 server

If you left because self-host is Enterprise-only: Langfuse

Galileo gates VPC and on-prem behind a contract. Langfuse is the opposite deal. It is MIT-licensed and self-hosts free, and only three features are enterprise-gated in the self-host build - tracing, evals, prompt management, human annotation and RBAC are all included. You run the real product on your own infrastructure, with your data never leaving your cloud, and you never talk to sales to do it. On the managed side it runs about $101/mo at 1M events, which is the cheapest way to stay hosted at real volume.

The catch is operational, not commercial. Langfuse v3 needs Postgres plus ClickHouse, Redis and S3-compatible storage - four services to stand up - and the migration to that architecture is where self-hosters get stuck, with GitHub threads reporting memory errors and timeouts. One more fact to file away - Langfuse was acquired by ClickHouse in January 2026. The MIT license and self-host survived, but you are now betting a multi-year stack on a database company’s subsidiary. Details are on the Langfuse page.

If you want transparent pricing on a card: Opik

Galileo’s pain is that ROI is unknowable without a call. Opik publishes everything. It is Comet’s open-source observability and eval platform, and its Pro cloud tier is $19/mo for 100k spans - the cheapest paid cloud of the major eval platforms, with overage at $5 per additional 100k. No sales motion, no quote. The OSS build is Apache-2.0, the most permissive license in this set, and Comet says the self-hosted version ships the full feature set with no gates and no caps. It is also the fastest-growing project of its peers at roughly 20.8k GitHub stars, and the eval side is real - LLM-as-judge, code metrics, online evaluation, guardrails and an Agent Optimizer.

The honest gotcha is per-seat pricing at scale. The recurring complaint is that seat costs climb as your team grows, so the $19 headline is the small-team configuration, not the large one. A few users report UI slowdown on very large projects. Neither undercuts the core value, but if you are a big team, model the seat cost before you commit.

If eval depth is the actual point: Arize Phoenix

A lot of teams look at Galileo for the eval metrics, not the platform. If that is you and you want it open and free, Arize Phoenix is the pick. It runs locally with a working trace UI in under a minute, it is built natively on OpenTelemetry and Arize’s OpenInference conventions so it is framework-agnostic, and its 50+ pre-built eval metrics include what reviewers call the best RAG evaluation in the category - serious retrieval and answer scoring without writing your own judge prompts. Arize is well-funded too, with a $70M Series C in February 2025, so it is not going anywhere.

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 are Apache-2.0. 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, but if you plan to resell it as a service, read the license first. There are also reports of ingest lag before traces appear. The Phoenix page walks through the nuance.

So which one?

• You are an enterprise that got burned by production hallucinations and can run a sales process - stay on Galileo. The Luna-plus-guardrails loop is a genuine story, and the free tier lets you evaluate it properly first. Just confirm you are looking at galileo.ai and not the Google-acquired design tool of the same name.
• You want free open-source self-host without a contract - Langfuse if you can run the four-service stack, Opik if you would rather stay managed at $19/mo.
• Eval depth, especially RAG, is the priority - Arize Phoenix, as long as you are not reselling it as a service.

One honest note on the numbers. The prices above are from each vendor’s public pages, verified on 26 July 2026, and this category ships breaking changes monthly - we re-verify every 30 days. And Galileo’s own headline Luna speed and cost claims are vendor-benchmarked, not independently confirmed, so treat them as a claim until you test on your own traffic. If open self-host or a price you can commit to is why you are here, the alternatives are not close.