List your company

Mathematical Reasoning on MATH Minerva

55.6ScoreICRL
40.62444.51248.452.288May 13, 2026

Evaluation results

MethodLinks
ICRLBackbone=Qwen3-8B2026.0555.6
GRPOBackbone=Qwen3-8B2026.0552.6
Critique-GRPOBackbone=Qwen3-8B2026.0552.6
SFTBackbone=Qwen3-8B2026.0543.8
Qwen3-8BBackbone=Qwen3-8B2026.0541.2