Mathematical Reasoning on GSM8K (Solve Rate)
94.4Solve RateAGENTCO-OP
Evaluation Results
| Method | Links | |
|---|---|---|
| AGENTCO-OPBackbone=GPT-4o-mini2026.05 | 94.4 | |
| ReConcileBackbone=GPT-4o-mini2026.05 | 93.7 | |
| AFlow*Backbone=Mixed2026.05 | 93.5 | |
| MultiPersonaBackbone=GPT-4o-mini2026.05 | 92.8 | |
| IOBackbone=GPT-4o-mini2026.05 | 92.7 | |
| CoT SCBackbone=GPT-4o-mini, Prompting Setting=5-shot2026.05 | 92.7 | |
| CoTBackbone=GPT-4o-mini2026.05 | 92.4 | |
| LLM-DebateBackbone=GPT-4o-mini2026.05 | 92.4 | |
| ADASBackbone=GPT-4o-mini2026.05 | 90.8 | |
| MedPromptBackbone=GPT-4o-mini2026.05 | 90 | |
| Self RefineBackbone=GPT-4o-mini2026.05 | 89.6 | |
| AFlowBackbone=GPT-4o-mini2026.05 | 86.8 | |
| Olmo3 BaseTuning Protocol=Freezing param.2026.07 | 37 | |
| LMK token tuningTuning Protocol=Freezing param.2026.07 | 37 | |
| HiLS-Attn HoPE-Q-CalTuning Protocol=Full param. CPT2026.07 | 36.85 | |
| HiLS-Attn RoPE-Q-CalTuning Protocol=Full param. CPT2026.07 | 35.71 | |
| HiLS-Attn NoPE-Q-CalTuning Protocol=Full param. CPT2026.07 | 35.03 | |
| Olmo3 512swaTuning Protocol=Full param. CPT2026.07 | 33.43 |