Long-CoT Reasoning on MMLU
90.3Accuracy (Long-CoT MMLU)SFI
Evaluation results
| Method | Links | |
|---|---|---|
| SFIModel=Qwen3-235B-A22B-Thinking-2507, Decoding Strategy=SFI (Ours)2026.03 | 90.3 | |
| SlowModel=Qwen3-235B-A22B-Thinking-2507, Decoding Strategy=Slow2026.03 | 90.09 | |
| SlowModel=Qwen3-30B-A3B-Thinking-2507, Decoding Strategy=Slow2026.03 | 70.9 | |
| SFIModel=Qwen3-30B-A3B-Thinking-2507, Decoding Strategy=SFI (Ours)2026.03 | 70.6 | |
| SlowModel=Qwen3-4B-Thinking-2507, Decoding Strategy=Slow2026.03 | 63 | |
| SFIModel=Qwen3-4B-Thinking-2507, Decoding Strategy=SFI (Ours)2026.03 | 63 |