Skip to content

Start free — no credit card required

Every plan runs real experiments with the same trustworthy results. Paid plans add automatic quality grading (LLM-as-judge), more team seats, more traffic, and longer data retention. Upgrade only when you outgrow Free.

LLMJury is in open beta. The statistics and SDKs are production-grade, but the product is young: expect rough edges, fast fixes, and frequent releases. Beta customers get a direct line to the founders — and pricing won’t change under you mid-subscription.

Free

$0/mo

Run your first real experiment and see the difference proof makes.

  • 50k events/mo
  • 5 seats
  • Latency, cost & business metrics
  • 7-day raw-event retention
  • All 3 SDKs (Python, TypeScript, Java)
  • Prompt versioning, history & rollback
  • Traffic-split safety check
  • Unlimited experiments
Start free

Pro

Most popular

$149/mo

+$49 per extra 500k events

For teams shipping LLM features who need automatic quality grading.

  • 1M events/mo
  • 100 seats
  • LLM-as-judge · 50k evaluations/mo
  • 30-day raw-event retention
  • Everything in Free
  • Custom judge metrics
  • Email support
Start free

Business

$799/mo

+$99 per extra 2M events

For production-scale products and larger teams.

  • 10M events/mo
  • Unlimited seats
  • LLM-as-judge · 500k evaluations/mo
  • 90-day raw-event retention
  • Everything in Pro
  • Method overrides
  • Priority support
Start free

Enterprise

Coming soon

Custom

Custom limits, security review, SSO, and a dedicated contact.

  • Custom allowance
  • Custom seats
  • Custom judge budget
  • Configurable retention
  • Custom limits
  • Security review
  • Dedicated support
Register interest

What the limits mean, in plain words

Events
One event = one thing you track (a model call, an exposure, or a business event like a conversion). Assigning a user to a variant is free and unlimited — it never counts.
Seats
How many teammates can join your organization and work on the same experiments. Free covers a small team (5 seats); Pro gives you 100, and Business is unlimited.
LLM-as-judge
Automatic quality grading: an LLM reads a sample of your outputs and scores them for quality, safety, and relevance — or against your own rubric. Judge grading runs real model calls, so it is included from Pro up with a generous monthly allowance. On Free you still measure latency, cost, and business outcomes.
Retention
How long we keep your raw event data. Your results are never deleted — finished results are saved permanently. Retention only limits how far back raw events can be re-analyzed.
Permutations & samples per arm
Ceilings on how much statistical compute each experiment can use. Defaults work for almost everyone; higher tiers let very large experiments run at higher precision.

Which plan am I?

One number decides it. Everything below is an estimate you can check — the arithmetic is on screen.

100k model calls = ~200k events a month — 2 per call, one for the exposure and one for the model call. Tracking a business outcome on every single request would take it to 300k; most teams track far fewer, because most requests are not conversions.

You want Pro — $149/mo, with 800k events of headroom.

An estimate, not a quote. Assignment itself is free and never counted, an experiment is always analyzed over its own full duration, and you can move plans at any time. If your integration looks different from the one above, ask us at [email protected].

Compare plans

Plan limit comparison
LimitFreeProBusinessEnterprise(coming soon)
Events50k events/mo1M events/mo10M events/moCustom allowance
Seats5 seats100 seatsUnlimited seatsCustom seats
Quality gradingLatency, cost & business metricsLLM-as-judge · 50k evaluations/moLLM-as-judge · 500k evaluations/moCustom judge budget
Retention7-day raw-event retention30-day raw-event retention90-day raw-event retentionConfigurable retention
Permutations1,000 permutations2,000 permutations10,000 permutationsConfigurable permutations
Samples per arm10k samples/arm50k samples/arm200k samples/armConfigurable samples/arm

Overage: Pro +$49 per extra 500k events; Business +$99 per extra 2M events. Finalized results are snapshotted, so they outlive raw-event retention. An experiment is always analyzed over its own full duration — no tier ever cuts an experiment short.

Coming soon

Enterprise

The Enterprise plan — custom limits, SSO, a security review, and a dedicated contact — is not available yet, and we want to build it around what you actually need. Where it sits relative to everything else is on the roadmap. Tell us about your requirements and we will contact you as soon as it opens (and you will help shape it).

Reviewing us before then? The security page covers what the SDK sends, encryption, key scoping, sub-processors, and exactly where we stand on certification.

or email us directly: [email protected]

Questions about the numbers above

Everything else — statistics, integration, what beta means — is in the full FAQ.
What counts as an event?
Three things: an exposure (a user was assigned to a variant), a model call the SDK measured, and a business event you sent with track(…). A single user interaction usually produces two or three. The allowance is per calendar month and resets at the start of the next one.
What happens when I hit my plan’s event allowance?
Your application keeps working. The SDK falls back to the in-code default prompt you passed, so nothing breaks on your request path, and nothing already stored is deleted — finalized results persist. On Pro and Business, overage is billed at $49 per extra 500k events and $99 per extra 2M respectively. On Free there is no overage: collection pauses until the month resets.
Does retention limit how long an experiment can run?
No. An experiment is always analyzed over its own full duration — the analysis window is never capped by tier. Retention is a separate, wider limit on how long raw events are kept: 7 days on Free, 30 on Pro, 90 on Business. If an experiment runs longer than its plan’s retention, only the retained tail of raw events can be re-analyzed, and the dashboard says so. Finalized results are snapshotted and kept regardless.
Why does quality grading start at Pro?
Judge evaluations are real model-inference calls with a real cost per evaluation, so they cannot be free. Free measures everything the SDK can capture without inference — latency, tokens, computed cost, and the business outcomes you send — which is enough to run a genuine experiment and read a statistically gated verdict.
What if I do not have enough traffic yet for a result to mean anything?
Two answers, and the first is the useful one. If you already have logged model calls and outcomes from before you adopted LLMJury, bulk-import them through POST /v1/events/import as CSV or JSON — original timestamps preserved — and you have a real baseline and the true distribution of your metrics on day one. What an import cannot do is manufacture a randomised comparison out of data that was never split, so the experiment still has to run. If you have no history either, prompt versioning, the audit trail, and rollback are all useful on the Free plan before a single result exists.
Do I need a credit card to start?
No. Every plan starts free with no card: all three SDKs, 50k events a month, unlimited experiments, and full prompt versioning. You add a card when you upgrade, and pricing is locked for existing subscribers during beta.