Evals
Eval results by metric and model, with score bars, regression marks and pass or review status. Fictitious data.
Evals
Quality scores for every dataset run
LatestLast 30 daysRun eval
Datasets□
14
Since last week▲ 2
Avg score□
92.7%
Since last week▲ 1.1 pts
Regressions□
1
Since last week▼ 1
Eval cost□
$38
Since last week▼ 6.0%
Latest run: support-answers v12240 cases
| Metric | Model | Score | vs last run | Distribution | Result |
|---|---|---|---|---|---|
| Answer accuracy | gpt-4o | 92.4% | +1.8 | Pass | |
| Answer accuracy | claude-sonnet | 94.1% | +0.6 | Pass | |
| Faithfulness | gpt-4o | 88.0% | -2.1 | Review | |
| Toxicity (lower is better) | gpt-4o | 0.4% | -0.1 | Pass | |
| JSON validity | gemini-pro | 99.2% | +0.0 | Pass |