Mathematical Reasoning on OlympiadBench (Avg@K)
44.25Avg Success Rate @8IntroLLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| IntroLLMBackbone=Qwen3-4B-Base2026.02 | 44.25 | — | — | |
| GRPOBackbone=Qwen3-4B-Base, Temperature (τ)=1.22026.02 | 44.12 | — | — | |
| EADBackbone=Qwen3-4B-Base, Schedule=1.2→0.12026.02 | 43.69 | — | — | |
| SENTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 42.85 | 42.99 | 42.95 | |
| w/ MaskBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 42.65 | 42.89 | 42.88 | |
| TAMPOBackbone=Qwen3-4B-Base, Range=0.6-1.52026.02 | 42.43 | — | — | |
| GRPOBackbone=Qwen3-4B-Base, Temperature (τ)=1.02026.02 | 42.36 | — | — | |
| w/ ClipBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 41.44 | 41.36 | 41.25 | |
| GRPOBackbone=Qwen3-4B-Base, Temperature (τ)=0.62026.02 | 41.06 | — | — | |
| w/ EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 40.48 | 40.11 | 40.39 | |
| w/ CovBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 39.72 | 40.29 | 40.92 | |
| w/ High EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 39.59 | 39.7 | 39.48 | |
| w/ AdvBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 37.04 | 36.28 | 36.24 | |
| Base modelBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 35.65 | 36.59 | 36.51 | |
| IntroLLMBackbone=Qwen3-1.7B-Base2026.02 | 29.95 | — | — | |
| GRPOBackbone=Qwen3-1.7B-Base, Temperature (τ)=1.22026.02 | 29.93 | — | — | |
| EADBackbone=Qwen3-1.7B-Base, Schedule=1.2→0.12026.02 | 29.64 | — | — | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 29.48 | 29.43 | 29.6 | |
| GRPOBackbone=Qwen3-1.7B-Base, Temperature (τ)=1.02026.02 | 28.78 | — | — | |
| GRPOBackbone=Qwen3-1.7B-Base, Temperature (τ)=0.62026.02 | 28.23 | — | — | |
| TAMPOBackbone=Qwen3-1.7B-Base, Range=0.6-1.52026.02 | 27.41 | — | — | |
| BaseBackbone=Qwen3-4B-Base2026.02 | 24.3 | — | — | |
| BaseBackbone=Qwen3-1.7B-Base2026.02 | 12.69 | — | — | |
| OpenPangu-embedded-1B-v1.1Base Model=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | — | 36.3 | — | |
| SENTBase Model=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | — | 42.99 | — |