Mathematical Reasoning on Gaokao (Pass@1, Pass@100)
64.7Pass@1MT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MTTeacher Model=Qwen2-Math-7B-Instruct2026.04 | 64.7 | — | |
| VCRD-Probapproach=PRM-free2026.04 | 64.2 | — | |
| distillm22026.04 | 61.6 | — | |
| DistillLM-2Teacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 54 | — | |
| VCRDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 53.2 | — | |
| GKDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 52.7 | — | |
| KDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 51.9 | — | |
| ABKDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 51.9 | — | |
| DistilLLMTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 50.9 | — | |
| LaDiRBackbone=LLAMA 3.1 8B2025.10 | 35.4 | 45.8 | |
| Discrete LatentBackbone=LLAMA 3.1 8B2025.10 | 33.3 | 39.7 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=0.12025.10 | 31.2 | 34.7 | |
| Soft ThinkBackbone=LLAMA 3.1 8B2025.10 | 31.2 | 33.4 | |
| TaH+Backbone=LLAMA 3.1 8B2025.10 | 30.9 | 34 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=0.72025.10 | 30.4 | 37.5 | |
| iCoTBackbone=LLAMA 3.1 8B2025.10 | 30 | 32.9 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=12025.10 | 29.7 | 39 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=1.22025.10 | 28.9 | 40.6 | |
| CODIBackbone=LLAMA 3.1 8B2025.10 | 27.5 | 35.2 | |
| CoconutBackbone=LLAMA 3.1 8B2025.10 | 26.8 | 29.3 | |
| LaDiR –w/o Stage 2Backbone=LLAMA 3.1 8B, Training Stage=without Stage 22025.10 | 24.7 | 30.2 | |
| Pause TokenBackbone=LLAMA 3.1 8B2025.10 | 21.3 | 25.7 | |
| CoT SFTBackbone=LLADA 8B2025.10 | 20.1 | 36 | |
| MSStudent Model=Qwen2-Math-1.5B2026.04 | 16.1 | — | |
| PLANNERBackbone=LLAMA 3.1 8B2025.10 | 15.1 | 21.3 | |
| LD4LGBackbone=LLAMA 3.1 8B2025.10 | 13.2 | 20.1 | |
| Base ModelBackbone=LLADA 8B2025.10 | 10.2 | 18.5 | |
| Sol-Only SFTBackbone=LLAMA 3.1 8B2025.10 | 0 | 2.8 |