Improvement
Side-by-side runs in Evals
Compare two prompts or models on the same test set, line by line.
Evals can now show two runs next to each other, so you can see exactly where a change made answers better or worse.
Scores, cost and response time sit beside each answer, and you can mark any line as a regression.
Saved comparisons become part of your release checklist, so nothing ships without a check.
