Mathematical Reasoning on OlympiadBench (Acc, TR, s/q)
63.1AccuracyFull CoT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full CoTBackbone=Llama-3.1-Nemotron-Nano-8B2026.05 | 63.1 | 0 | 2.83 | |
| PUMABackbone=Llama-3.1-Nemotron-Nano-8B2026.05 | 62.2 | 27 | 2.3 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 62.1 | 9.5 | 10.36 | |
| DynasorBackbone=Llama-3.1-Nemotron-Nano-8B2026.05 | 61.9 | -3 | 24.32 | |
| Full CoTBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 61.5 | 0 | 7.32 | |
| DEERBackbone=Llama-3.1-Nemotron-Nano-8B2026.05 | 61.2 | 12.4 | 4.99 | |
| DynasorBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 60.1 | 11.4 | 42.96 | |
| PUMABackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 59.6 | 31.9 | 5.39 | |
| Full CoTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.05 | 57.6 | 0 | 2.57 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.05 | 57.5 | 20.2 | 3.78 | |
| PUMABackbone=DeepSeek-R1-Distill-Qwen-7B2026.05 | 55.6 | 43.2 | 1.64 | |
| DynasorBackbone=DeepSeek-R1-Distill-Qwen-7B2026.05 | 49 | 34.1 | 1.84 |