List your company
ResearchFormerly WizWand

Mathematical Reasoning on Concatenated Mixed-Difficulty Benchmark

92.2AccuracyZIP-RC sampling
66.82473.4128086.588Dec 1, 2025
Updated 8mo ago

Evaluation results

MethodLinks
ZIP-RC samplingModel=Qwen3-1.7B2025.1292.21.43
Weighted BoN Self-evalModel=Qwen3-1.7B2025.1291.61.4
Majority VotingModel=Qwen3-1.7B2025.12911.4
Weighted BoN ext. RMModel=Qwen3-1.7B2025.12911.43
ZIP-RC reward pruneModel=Qwen3-1.7B2025.1288.91.33
MV length-pruneModel=Qwen3-1.7B2025.12881.46
ZIP-RC samplingModel=LFM2-1.2B2025.12861.35
ZIP-RC reward pruneModel=LFM2-1.2B2025.1285.81.49
MV length-pruneModel=LFM2-1.2B2025.1284.91.7
Weighted BoN Self-evalModel=LFM2-1.2B2025.1284.91.6
Majority VotingModel=LFM2-1.2B2025.1283.81.6
Weighted BoN ext. RMModel=LFM2-1.2B2025.1282.51.53
ZIP-RC samplingModel=LFM2-350M2025.1274.11.49
Weighted BoN Self-evalModel=LFM2-350M2025.1271.11.7
MV length-pruneModel=LFM2-350M2025.1270.61.66
Majority VotingModel=LFM2-350M2025.1268.81.7
Weighted BoN ext. RMModel=LFM2-350M2025.1267.81.59
ZIP-RC reward pruneModel=LFM2-350M2025.1267.81.27