MATH (Hendrycks et al.) — Explicit CoT+2-shot axis (Ministral 3, 0-1 scale); distinct from other MATH variants.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Ministral 3 14B | Mistral | 67.6 | 4 | 2026-06-12 |
| 2 | Ministral 3 8B | Mistral | 62.6 | 3 | 2026-06-05 |
| 3 | Qwen3 14B Base | Alibaba | 62 | 4 | 2026-06-05 |
| 4 | Ministral 3 3B | Mistral | 60.1 | 3 | 2026-06-05 |
| 5 | Qwen3 8B Base | Alibaba | 57.6 | 4 | 2026-06-05 |
| 6 | Qwen3 4B Base | Alibaba | 40.5 | 4 | 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.