OpenAI Evals is shutting down. Take your evals with you.

OpenAI has announced that the Evals platform becomes read-only on 31 October 2026 and shuts down on 30 November 2026. If your eval data is a JSONL file, it doesn't have to go with it: GetEvals reads OpenAI Evals data files directly, discovers the failure modes your data actually contains, and hands back a suite that runs anywhere.

Drop your traces here

JSONL · JSON · NDJSON · CSV · TSV · TXT · GZ · ZIP

Moving your data across

  1. 1Download your eval data file from the platform, or from the Files API with purpose: "evals".
  2. 2Drop it above. The {"item": {...}} wrapper is understood, and so are sample blocks holding the model output that was graded.
  3. 3Any label, correct, passed, or score field is read as ground truth — so the report can tell you how well an automated judge matches your existing labels.
  4. 4Download the exports: a fresh OpenAI Evals JSONL, a promptfoo config, and a portable judge-prompts.json that runs in any harness.

Trace grading is covered too: if you export the traces behind an eval, the analysis works on the full trajectory including tool calls, not just the final answer.

What you get back

A self-contained HTML report: the failure modes discovered in your own traces (not a generic checklist), a heatmap of which traces hit which mode, the exact quoted evidence behind every verdict, and the judge's own reliability numbers. Plus downloadable evals — a promptfoo config, an OpenAI Evals data file, and a portable judge suite — so the analysis turns into a regression suite you keep.

Related