List your company

Mathematical Reasoning on HMMT-F26 (avg@16, pass@16)

16.29Avg@16Rollout-level |Ai| replay
-0.4543.8938.2412.587Jun 3, 2026

Evaluation results

MethodLinks
Rollout-level |Ai| replayModel=Qwen3-4B2026.0616.2930.3
σgModel=Qwen3-4B2026.0611.3627.27
uniformModel=Qwen3-4B2026.0610.4426.38
GRPOModel=Qwen3-4B2026.0610.2321.21
GRPOModel=Qwen3-1.7B2026.063.229.09
σgModel=Qwen3-1.7B2026.063.0312.12
Rollout-level |Ai| replayModel=Qwen3-1.7B2026.062.8415.15
uniformModel=Qwen3-1.7B2026.061.789.09
Rollout-level |Ai| replayModel=Qwen3-0.6B2026.061.3312.12
GRPOModel=Qwen3-0.6B2026.060.769.09
σgModel=Qwen3-0.6B2026.060.7612.12
uniformModel=Qwen3-0.6B2026.060.193.03