Judgment Labs pricing

What Is Published and What Is Not

Judgment Labs has no public price. judgmentlabs.ai/pricing returns a 404. The site's sitemap lists no pricing or plans page, while the homepage sends prospective buyers to "Book demo". The docs explain traces, judges, behaviors, datasets, tests and monitoring without describing plans.

An unpublished price says nothing about the number. It says how the company sells. Judgment Labs' May 2026 funding announcement says it is building a forward-deployed engineering team, and a demo-led sale with hands-on onboarding fits that model. Treat any Judgment Labs figure you find outside a sales conversation as a guess.

What You Can Use Without Paying

The judgeval Python SDK is open source under Apache-2.0, with JavaScript, Go and Java siblings. You can install it, inspect it, fork it and ship it inside a commercial product without paying for the SDK.

Free to install is not free to run. The README's quickstart begins by exporting a JUDGMENT_API_KEY and a JUDGMENT_ORG_ID, and its online monitoring scores "live production traffic server-side" - on the Judgment platform. Trace storage, judges on live traffic, dashboards and alerts are platform features, and the platform is what the demo prices. Whether new accounts get a free allowance is not published. More on the split in is judgeval open source.

What to Ask on the Demo

Go in with your own numbers - sessions a day, average trace size, agents, environments, people who need access - and ask for the answers in writing. These questions make a quote comparable with a published price list:

1. What is the meter? Traces ingested, spans, judge evaluations, seats, a platform fee, or a mix. The meter decides how the bill grows with traffic more than the first-year number does.
2. Who pays for judge tokens? The judges docs do not say which model Agent Judge runs on or whether you can bring your own key. Ask both, and whether judge runs are capped.
3. What share of traffic gets judged? The monitoring docs say scores cover "traces the judge evaluated, not necessarily every trace sent to the project". Ask what sampling your price includes and what full coverage costs.
4. How long is data kept? Retention for traces and for scores, and the cost of keeping either longer.
5. Where are SSO, SAML and RBAC? Which plan includes them, and whether they require an enterprise agreement.
6. When does self-hosting arrive, and at what price? The docs say coming soon. If you need it, get the date and the terms into the contract.
7. Can you export everything? Traces, scores, rubrics and datasets, in what format, if you leave.
8. What are the contract terms? Minimum term, annual or monthly billing, and whether a paid pilot is possible before a commitment.

What the Rest of the Category Publishes

Every other agent evaluation vendor in this comparison publishes at least an entry tier:

Read it with two cautions. The meters do not convert: a run, an event, a trace, a credit, a unit, a gigabyte and a score are different things, and Latitude does not say what a credit measures. And every vendor's enterprise tier is quoted, Failproof AI's included - published pricing matters most below that line, which is where most 50-to-500-person teams buy. Agent observability pricing compared works through the arithmetic at real volumes.

The Second Bill: Judge Tokens

An LLM-based evaluation also consumes model tokens. The platform may include that usage, meter it separately or require your own model-provider key. Ask which applies before comparing quotes. Failproof AI includes 100 evaluations on Free, 50,000 on Team and 500,000 on Scale. Each session-and-evaluation pair counts once, with evaluation overage at $0.05 on Team and $0.04 on Scale. Cloud runs the code-based and LLM-based evaluations, including your existing evaluation suite.

For example, assuming one evaluation per session and no run overage, 10,000 or 30,000 evaluated sessions both fit inside Team for $99. At 75,000, Team is $99 + 25,000 × $0.05 = $1,349, while Scale covers the evaluations for $599. Model-token charges, when separate, come on top. Reduce those by using code checks for exact conditions and reserving model-based evaluations for criteria that require judgment.

When the Missing Price Does Not Matter

If you want what Judgment Labs is built around - hosted judges with citations to the exact span, calibration tooling and rubric-rewrite suggestions, from a research-focused team - and your company buys software through demos and annual contracts anyway, the missing price list costs you a call, not a decision. Book it, bring the questions above, and compare the quote on the same meter as the alternatives. If you need to start this week on a card, or you need a bad action blocked rather than scored, look at the alternatives in parallel; nothing stops you running a self-serve trial while the quote is prepared.