Opik Review (2026)

What Opik actually is

Opik is Comet’s open-source LLM observability and evaluation platform. It captures multi-step traces - LLM calls, tool executions, agent activity - and layers on datasets, experiments, code-based and LLM-as-judge metrics, online evaluation, Guardrails and an Agent Optimizer. In one line, it is tracing plus eval plus prompt management plus production monitoring, in a single open-source package.

What makes it stand out is not any one feature - the category is crowded with tracing-plus-eval tools. It is the terms. Opik has the most permissive OSS license of the major platforms and the cheapest paid cloud, and it is the fastest-growing project of the bunch. Those three facts, together, are the story.

It is built by Comet, an established ML platform company founded in 2017 by Gideon Mendels and Nimrod Lahav, headquartered in New York. Comet raised a $50M Series B in November 2021 led by OpenView, roughly $63M in total, and launched Opik on 17 September 2024 under Apache-2.0. So this is not a solo maintainer’s weekend project - it is a mature company’s open-source bet, which matters when you are choosing a platform to depend on.

The reason to pick it: open source without the catch

This goes first because it is the whole differentiator. A lot of “open-source” observability tools have a catch - a restrictive license, or a hobbled core with the good features gated behind a paid cloud. Opik’s OSS build is Apache-2.0, and Comet states it ships the full feature set self-hosted, with unlimited spans, members and retention, and no feature gates.

Contrast that with the closest comparisons. Phoenix’s server is Elastic License 2.0 - source-available, not OSI open source, and it forbids offering it as a hosted service. Opik is Apache-2.0, a genuinely permissive OSI license with none of that restriction. And it is designed to self-host at Kubernetes scale without losing features, so the self-hosted product is the real product. If your requirement is “open source that actually means open source, and I keep every feature when I self-host,” Opik is the cleanest answer in the category.

The market seems to agree. Opik is at roughly 20.8k GitHub stars and is the fastest-growing of its peers - it went from zero to 12.5k stars in about eight or nine months. Reviewers praise the dashboard quality and the depth of experiment tracking Comet brings from its ML-platform heritage.

Pricing: the cheapest cloud in the category

Tier	Price	Spans	Members	Retention
Open Source	$0	unlimited	unlimited	unlimited
Free Cloud	$0	25k/mo	up to 10	60 days
Pro Cloud	$19/mo	100k/mo	up to 50	60 days
Enterprise	Contact sales	unlimited	unlimited	custom

The self-hosted OSS build is free with no caps. On the cloud, the Free tier gives you 25k spans/mo, up to 10 members and 60-day retention.

The number worth underlining: Pro Cloud is $19/mo for 100k spans - the cheapest paid cloud tier of the major eval platforms. Overage is $5 per additional 100k spans, and you can extend retention to 400 days at $29 per 100k spans. Compare that to Confident AI’s Starter at $200/mo and the gap is an order of magnitude. If price is the deciding factor and you want managed hosting, Opik is hard to beat.

The quota unit is spans per month plus members. An annual discount is not stated on the pricing page.

The gotcha: per-seat pricing at scale

The honest weakness is the pricing shape as your team grows. The recurring complaint is that per-seat pricing scales poorly for larger teams - smaller startups report that the cost gets hard to justify as headcount rises, even though the entry price is low. A few users also report occasional UI slowdown on very large projects. Neither undercuts the core value proposition, but if you are a big team eyeing the cloud, model the seat cost before you commit - the $19 headline is per the small-team configuration, not the large one.

Opik versus Phoenix

The two open-source observability defaults, and the split is about license and architecture. Opik is Apache-2.0 with a full-featured self-host and the cheapest cloud; Phoenix is Elastic License 2.0 server with deeper RAG eval and a stronger OTel-native story. If your priority is a permissive license and a cheap managed option, Opik wins. If your priority is the deepest RAG evaluation and strict OpenTelemetry-native architecture, Phoenix has the edge - Opik treats OTel as one ingestion path rather than the core.

For most teams choosing an open-source platform to self-host without a license asterisk, Opik is the safer default. For eval depth specifically, look hard at Phoenix.

Should you use it?

Use Opik if you want a genuinely feature-complete open-source observability platform you can self-host under Apache-2.0 with no gates, or you want the cheapest managed cloud in the category, or you want tracing and eval from a company with real staying power.

Think twice if you are a large team sensitive to per-seat cloud pricing - model it first - or if you need strict OpenTelemetry-native architecture, where Phoenix is the better fit. Neither is a dealbreaker for most buyers; both are worth knowing.

Bottom line: Opik is the most honest open-source deal in the category. The license is permissive, the self-host is genuinely complete, and the cloud is the cheapest option here. The growth curve is not an accident - it is the market noticing exactly that.

---

Pricing and features verified against comet.com on 23 July 2026. This category ships breaking changes monthly - we re-verify every 30 days.