Mathematical Reasoning on Math500, AIME24, AMC, Gaokao, Minerva Math, and Olympiad Bench
93.2Math500 ScoreGRESO
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GRESOBackbone=Qwen2.5-Math-7B, #Rollout=6.3M (2.1×)2026.03 | 93.2 | 30 | 78.9 | 70.2 | 35.2 | 44.1 | 58.6 | |
| HIVEBackbone=Qwen2.5-Math-7B, #Rollout=3.9M (3.4×)2026.03 | 93 | 36.7 | 79.2 | 70.8 | 34.8 | 43.8 | 59.7 | |
| DSBackbone=DeepSeek-R1-Distill-Qwen-1.5B, #Rollout=2.4M (1.0×)2026.03 | 87.9 | 36.7 | 71.7 | 78.7 | 35.3 | 54.9 | 60.9 | |
| HIVEBackbone=DeepSeek-R1-Distill-Qwen-1.5B, #Rollout=1.5M (1.6×)2026.03 | 87.2 | 37.5 | 70.1 | 77.2 | 35.1 | 53.2 | 60.1 | |
| GRESOBackbone=DeepSeek-R1-Distill-Qwen-1.5B, #Rollout=1.6M (1.5×)2026.03 | 85.5 | 37.5 | 70.7 | 76.2 | 34.3 | 52.1 | 59.4 | |
| DSBackbone=Qwen2.5-Math-7B, #Rollout=13.1M (1.0×)2026.03 | 82.9 | 34.2 | 79.2 | 71.7 | 35.4 | 43.6 | 57.8 | |
| HIVEBackbone=Qwen2.5-Math-1.5B, #Rollout=3.1M (2.5×)2026.03 | 77.9 | 16.7 | 60.2 | 66.1 | 31.5 | 40.3 | 48.8 | |
| DSBackbone=Qwen2.5-Math-1.5B, #Rollout=7.6M (1.0×)2026.03 | 77.3 | 16.7 | 61.7 | 64.2 | 31.8 | 38.7 | 48.4 | |
| GRESOBackbone=Qwen2.5-Math-1.5B, #Rollout=3.3M (2.3×)2026.03 | 77.3 | 15 | 59.3 | 66.2 | 32.6 | 38.5 | 48.1 | |
| HIVEBackbone=Llama3.2-3b-Instruct, #Rollout=4.8M (2.4×)2026.03 | 53.8 | 17.5 | 45.5 | 49.1 | 20.5 | 20 | 34.4 | |
| DSBackbone=Llama3.2-3b-Instruct, #Rollout=11.4M (1.0×)2026.03 | 52.4 | 16.7 | 46 | 49.2 | 20 | 20.2 | 34.1 | |
| GRESOBackbone=Llama3.2-3b-Instruct, #Rollout=5.9M (1.9×)2026.03 | 51.8 | 16.7 | 46.3 | 49.4 | 19.7 | 19.6 | 33.9 |