Simple eval pricing that scales with your deployment
Priced per eval run — the unit your CI/CD pipeline already thinks in. A team running 5 features with 20-criteria eval suites at 10 deploys/week consumes roughly 10,000 eval runs/month. Start free and move to Team when your deploy cadence outgrows that.
- 10,000 eval runs/month
- 3 eval criteria types
- Prompt version tracking
- 7-day result retention
- Community support
- 200,000 eval runs/month
- All eval criteria types
- CI/CD integration (GitHub, GitLab, Vercel)
- Custom LLM-as-judge criteria
- Regression alerting via Slack/email
- 90-day result retention
- Email support
- Unlimited eval runs
- Custom eval pipeline integration
- SSO / SAML
- Dedicated Slack channel
- Eval data residency options
- SLA + uptime guarantee
- Onboarding + quarterly reviews
Annual price is per-month billed annually. Monthly billing available on Team tier. Overage on Starter: $0.004 per additional eval run.
Pricing questions
One eval run = one LLM call evaluated against one criterion. If your eval suite has 10 criteria and you run 100 LLM calls, that's 1,000 eval runs. Fyntune shows your run consumption on the dashboard in real time.
On Starter, evals above 10,000/mo are billed at $0.004 per run. On Team, overages are $0.002 per run. We send alerts at 80% and 100% of your included limit. You can set a monthly overage cap if you prefer a hard cutoff.
A seat is any user with access to the Fyntune dashboard or API. Read-only collaborators count as seats. You can add or remove seats at any time; billing is prorated to the day.
Starter: 7 days. Team: 90 days. Enterprise: configurable (up to unlimited). Note that raw LLM output data is retained for a maximum of 30 days across all tiers — after that, only aggregate eval scores and deltas are kept. Enterprise customers can configure custom data residency and retention terms.
Catching one regression is worth the subscription
One production LLM regression — a factuality spike, a guardrail bypass, a hallucination pattern — that requires two engineer-hours to diagnose and fix costs more than a month of Fyntune Team. Fyntune doesn't replace your observability stack. It fills the specific gap between "tests passed in CI" and "users are reporting quality issues."