List your company

Mathematical Reasoning on GSM8K (Random, Reward, ∆ (pp))

90.4Random BaselineSparse Reward
36.3250.3664.478.44Oct 2, 2025

Evaluation results

MethodLinks
Sparse RewardBackbone=Qwen3-4B2025.1090.4932.5
Dense RewardBackbone=Qwen3-4B2025.1089.691.41.8
Interval RewardBackbone=Qwen3-4B2025.1087.891.23.5
Sparse RewardBackbone=Qwen2.5-7B2025.1085.888.83
Sparse RewardBackbone=Llama3.1-8B2025.1080.682.11.5
Interval RewardBackbone=Qwen2.5-7B2025.1078.882.53.7
Interval RewardBackbone=Llama3.1-8B2025.1067.678.210.7
Dense RewardBackbone=Llama3.1-8B2025.1064.671.16.5
Dense RewardBackbone=Qwen2.5-7B2025.1038.455.717.4