Mathematical and Scientific Reasoning on OlympiadBench
50.24Mean AccuracyINFUSER
Evaluation results
| Method | Links | |
|---|---|---|
| INFUSERModel Backbone=Qwen3-8B-Base2026.06 | 50.24 | |
| AZRModel Backbone=Qwen3-8B-Base2026.06 | 47.92 | |
| R-FewModel Backbone=Qwen3-8B-Base2026.06 | 46.4 | |
| R-ZeroModel Backbone=Qwen3-8B-Base2026.06 | 45.1 | |
| Qwen2.5 32B + RL (synthetic)Model Size=32B, Post-training=RL (synthetic)2026.04 | 44.53 | |
| R-ZeroModel Backbone=Qwen3-4B-Base2026.06 | 43.18 | |
| AZRModel Backbone=Qwen3-4B-Base2026.06 | 43.18 | |
| R-FewModel Backbone=Qwen3-4B-Base2026.06 | 42.8 | |
| SPICEModel Backbone=Qwen3-4B-Base2026.06 | 42.7 | |
| SPICEModel Backbone=Qwen3-8B-Base2026.06 | 42.5 | |
| INFUSERModel Backbone=Qwen3-4B-Base2026.06 | 42.38 | |
| Qwen2.5 32BModel Size=32B, Variant=Instruct2026.04 | 41.41 | |
| BaseModel Backbone=Qwen3-8B-Base2026.06 | 40.36 | |
| BaseModel Backbone=Qwen3-4B-Base2026.06 | 35.31 |