MMLU-Redux — Same real benchmark as mmlu_redux, but explicit 5-shot+0-1 scale+Ministral roster kept separate.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen3 14B Base | Alibaba | 83.7 | 4 | 2026-06-05 |
| 2 | Ministral 3 14B | Mistral | 82 | 3 | 2026-06-05 |
| 3 | Qwen3 8B Base | Alibaba | 79.4 | 4 | 2026-06-05 |
| 4 | Ministral 3 8B | Mistral | 79.3 | 3 | 2026-06-05 |
| 5 | Qwen3 4B Base | Alibaba | 75.9 | 4 | 2026-06-05 |
| 6 | Ministral 3 3B | Mistral | 73.5 | 3 | 2026-06-05 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.