Mathematical Reasoning on AIME In-Distribution 2026 (Accuracy)
70AccuracySkillFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillFlowMethod Category=Ours2026.05 | 70 | |
| FlowSteerMethod Category=Agent+RL2026.05 | 63.33 | |
| AgentFlowMethod Category=Agent+RL2026.05 | 60 | |
| SkillRLMethod Category=Agent+RL2026.05 | 56.67 | |
| Qwen3.5-AFlowMethod Category=AFlow, Backbone=Qwen3.52026.05 | 53.33 | |
| v4-flashMethod Category=Baseline2026.05 | 50 | |
| Qwen3.5-9BMethod Category=Baseline2026.05 | 46.67 | |
| Qwen3.5-SFTMethod Category=SFT, Backbone=Qwen3.52026.05 | 36.67 | |
| Qwen3.5-GRPOMethod Category=GRPO, Backbone=Qwen3.52026.05 | 33.33 |