List your company

Mathematical Reasoning on MATH500 (% Avg@4)

94.1Avg@4 (%)GRPO + RePro
75.69280.47185.2590.029Dec 1, 2025

Evaluation results

MethodLinks
GRPO + ReProBackbone=Qwen3-1.7B2025.1294.1
GRPOBackbone=Qwen3-1.7B2025.1293.4
RF++ B + ReProBackbone=Qwen3-1.7B2025.1292.7
PPO + ReProBackbone=Qwen3-1.7B2025.1292.4
PPOBackbone=Qwen3-1.7B2025.1292.1
OriginalBackbone=Qwen3-1.7B2025.1291.5
RF++ BBackbone=Qwen3-1.7B2025.1291.5
RF++ B + ReProBackbone=Hunyuan-1.8B-Instruct2025.1286.2
PPOBackbone=Hunyuan-1.8B-Instruct2025.1286
GRPOBackbone=Hunyuan-1.8B-Instruct2025.1285.6
RF++ BBackbone=Hunyuan-1.8B-Instruct2025.1285.5
PPO + ReProBackbone=Hunyuan-1.8B-Instruct2025.1284.3
GRPO + ReProBackbone=Hunyuan-1.8B-Instruct2025.1284.2
PPO + ReProBackbone=MobileLLM-R1-950M2025.1283.5
PPOBackbone=MobileLLM-R1-950M2025.1281.4
OriginalBackbone=Hunyuan-1.8B-Instruct2025.1281.3
OriginalBackbone=MobileLLM-R1-950M2025.1276.4