MRCR v2 (OpenAI long-context needle benchmark) — v2, 8-needle, 1M context (distinct length), pointwise metric - both axes distinct.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Gemini 3 Pro | 26.3 | 1 | 2026-05-01 | |
| 2 | Gemini 3 Flash Preview | 26.3 | 1 | 2026-07-06 | |
| 3 | Gemini 3.1 Pro | 26.3 | 1 | 2026-05-01 | |
| 4 | Gemini 3.5 Flash | 22.1 | 1 | 2026-07-06 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.