Mathematical Reasoning on MATH-500 (ACC, A.Tok, ε³)
89.9AccuracyR1-VeriThinker
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| R1-VeriThinkerBackbone=DeepSeek-R1-Distill-Llama-8B2026.05 | 89.9 | 2,953 | 2.74 | |
| VPG-EABackbone=DeepSeek-R1-Distill-Llama-8B2026.05 | 89.6 | 2,536.2 | 3.17 | |
| Manifold SteeringBackbone=DeepSeek-R1-Distill-Llama-8B2026.05 | 88 | 2,873 | 2.7 | |
| Original ModelBackbone=DeepSeek-R1-Distill-Llama-8B2026.05 | 87.4 | 3,292.8 | 2.32 | |
| ACPOBackbone=DeepSeek-R1-Distill-Llama-8B2026.05 | 87 | 2,232 | 3.39 |