Question Answering on SimpleQA (Accuracy)
6.73AccuracyMARGO
Evaluation Results
| Method | Links | |
|---|---|---|
| MARGOBackbone=Qwen3-4B2026.07 | 6.73 | |
| MARGOBackbone=Qwen3-8B2026.07 | 5.82 | |
| FixedThink+RLBackbone=Qwen3-8B2026.07 | 5.29 | |
| AdaptiveRLBackbone=Qwen3-8B2026.07 | 5.27 | |
| FixedThinkBackbone=Qwen3-8B2026.07 | 5.04 | |
| FixedNoThink+RLBackbone=Qwen3-4B2026.07 | 4.62 | |
| AdaptiveCLSBackbone=Qwen3-8B2026.07 | 4.62 | |
| FixedThink+RLBackbone=Qwen3-4B2026.07 | 4.48 | |
| AdaptiveSFTBackbone=Qwen3-8B2026.07 | 4.3 | |
| AdaptiveRLBackbone=Qwen3-4B2026.07 | 4.23 | |
| FixedNoThink+RLBackbone=Qwen3-8B2026.07 | 4.23 | |
| FixedNoThinkBackbone=Qwen3-8B2026.07 | 4.16 | |
| FixedThinkBackbone=Qwen3-4B2026.07 | 3.98 | |
| FixedNoThinkBackbone=Qwen3-4B2026.07 | 3.74 | |
| AdaptiveCLSBackbone=Qwen3-4B2026.07 | 3.61 | |
| AdaptiveSFTBackbone=Qwen3-4B2026.07 | 3.35 |