Scientific Reasoning on OlympiadBench
76.4Average ScoreOuro-2.6B-Thinking-R4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ouro-2.6B-Thinking-R4Parameters=2.6B, Configuration=Thinking-R42025.10 | 76.4 | — | — | |
| Qwen3-8BParameters=8B2025.10 | 75.3 | — | — | |
| Qwen3-4BParameters=4B2025.10 | 73.2 | — | — | |
| Deepseek-Distill-Qwen-7BParameters=7B2025.10 | 72 | — | — | |
| Ouro-1.4B-Thinking-R4Parameters=1.4B, Configuration=Thinking-R42025.10 | 71.6 | — | — | |
| Deepseek-Distill-Qwen-1.5BParameters=1.5B2025.10 | 56.44 | — | — | |
| Qwen3-1.7BParameters=1.7B2025.10 | 56.4 | — | — | |
| TCA-AttentionBase Model=Qwen2.5-7B-Instruct, Attention Method=TCA-Attention2025.12 | 31.27 | 38.82 | 19.95 | |
| Qwen2.5-7B-InstructBase Model=Qwen2.5-7B-Instruct, Attention Method=Full Attention2025.12 | 31.2 | 38.85 | 19.73 | |
| MInferenceBase Model=Qwen2.5-7B-Instruct, Attention Method=MInference2025.12 | 30.99 | 38.79 | 19.28 | |
| XAttentionBase Model=Qwen2.5-7B-Instruct, Attention Method=XAttention2025.12 | 30.9 | 38.94 | 18.84 | |
| FlexPrefillBase Model=Qwen2.5-7B-Instruct, Attention Method=FlexPrefill2025.12 | 30.86 | 38.73 | 19.06 | |
| TCA-AttentionBase Model=LLaMA3.1-8B-Instruct, Attention Method=TCA-Attention2025.12 | 12.15 | 13.19 | 10.59 | |
| MInferenceBase Model=LLaMA3.1-8B-Instruct, Attention Method=MInference2025.12 | 12.08 | 13.64 | 9.73 | |
| XAttentionBase Model=LLaMA3.1-8B-Instruct, Attention Method=XAttention2025.12 | 11.95 | 13.71 | 9.3 | |
| LLaMA3.1-8B-InstructBase Model=LLaMA3.1-8B-Instruct, Attention Method=Full Attention2025.12 | 11.84 | 13.84 | 8.85 | |
| FlexPrefillBase Model=LLaMA3.1-8B-Instruct, Attention Method=FlexPrefill2025.12 | 11.57 | 13.39 | 8.85 |