Facet Evals
LayerEvals
p95 verdict240 ms
CoveragePrompt & Agent
ComplianceSOC 2 Type II

Facet

Evals © '26
Run 4,182 · click to resample
Start free

Gating releases at

Evals for
shipping teams

Release gate · v4

An eval suite your model has to pass before it reaches anyone

Graded, not vibed

Every release runs the same graded suite, on the same fixtures, with the diff attached.

Read the method
Latest

Run gallery

4,182 runs graded this quarter

The gate, as your reviewer sees it

Every pull request gets a verdict

Facet grades the candidate prompt against the baseline on the same fixtures and posts the diff to the PR. A regression past threshold blocks the merge, with the failing cases attached.

Open a sample report
facet / support-agent / PR #2814 Gate blocked · 3 regressions
Baseline v4.2Candidate v4.31,240 cases3 graders4m 12s
SuiteBaselineCandidateChange
Groundedness · last 20 runs
Case 0417 · groundedness · score 0.21
In
Can I still return a jacket I bought 45 days ago?
Out
Yes, returns are accepted within 60 days of purchase.
Judge
Contradicts retrieved policy §2.1 (30 days). No source cited.

What teams say

Receipts, not adjectives

Each quote comes with the number behind it, pulled from the team's own workspace.

From their workspace

Pricing

Priced per graded case: one fixture, run through every grader in the suite.

Starter

$0/ month

  • 1,000 graded cases
  • 2 suites
  • Model-as-judge with your own key
  • Community support
Start free

Team

$490/ month

  • 50,000 graded cases, then $0.008 each
  • Unlimited suites
  • PR gates with failing-case reports
  • Trace scoring on 1M spans
  • Slack and email drift alerts
Start a 14-day trial

Enterprise

Custom

  • Volume pricing on cases
  • SSO, SCIM and audit log
  • Deploy in your VPC
  • SOC 2 Type II report
  • Named support engineer
Talk to us
Start free

Get started

Run your first suite

Create a workspace

npx facet init

docs.facet.dev

team@facet.dev