List your company

Mathematical Reasoning on GSM8K (test) (Accuracy, Tokens, Calls, Agents)

92.4AccuracySelf-Consistency
68.68874.8448187.156May 27, 2026

Evaluation results

MethodLinks
Self-ConsistencyBackbone=Qwen2.5-7B-Instruct2026.0592.413,551.4101
CoTBackbone=Qwen2.5-7B-Instruct2026.0591.61,355.111
Self-ConsistencyBackbone=Phi-3 Mini 4K Instruct2026.0590.3714,610.8101
TRACERBackbone=Qwen2.5-7B-Instruct2026.0589.011,014.13.022
Sparse MADBackbone=Qwen2.5-7B-Instruct2026.0588.785,488.163
MoABackbone=Qwen2.5-7B-Instruct2026.0588.78,408.5121
MADBackbone=Qwen2.5-7B-Instruct2026.0588.625,909.163
Single-Agent GRPOBackbone=Qwen2.5-7B-Instruct2026.0588.25252.311
TRACERBackbone=Phi-3 Mini 4K Instruct2026.0588.02959.72.942
Single-Agent GSPOBackbone=Phi-3 Mini 4K Instruct2026.0587.95276.211
Single-Agent GSPOBackbone=Qwen2.5-7B-Instruct2026.0587.41246.211
Single-Agent GRPOBackbone=Phi-3 Mini 4K Instruct2026.0586.81282.111
Sparse MADBackbone=Phi-3 Mini 4K Instruct2026.0584.694,95763
MAPoRLBackbone=Phi-3 Mini 4K Instruct2026.0584.151,770.742
MADBackbone=Phi-3 Mini 4K Instruct2026.0584.085,34563
MAPoRLBackbone=Qwen2.5-7B-Instruct2026.0583.852,098.942
CoTBackbone=Phi-3 Mini 4K Instruct2026.0582.51,461.111
MoABackbone=Phi-3 Mini 4K Instruct2026.0580.447,995.4121
MAGRPOBackbone=Phi-3 Mini 4K Instruct2026.0572.521,267.142
MAGRPOBackbone=Qwen2.5-7B-Instruct2026.0569.61,29742