Question Answering on SimpleQA-V
10AccuracyMARGO
Evaluation Results
| Method | Links | |
|---|---|---|
| MARGOBackbone=Qwen3-4B2026.07 | 10 | |
| MARGOBackbone=Qwen3-8B2026.07 | 9 | |
| FixedNoThink+RLBackbone=Qwen3-8B2026.07 | 5.6 | |
| AdaptiveRLBackbone=Qwen3-8B2026.07 | 5.5 | |
| FixedNoThink+RLBackbone=Qwen3-4B2026.07 | 5.2 | |
| FixedThink+RLBackbone=Qwen3-8B2026.07 | 4.9 | |
| AdaptiveSFTBackbone=Qwen3-4B2026.07 | 4.8 | |
| FixedThinkBackbone=Qwen3-8B2026.07 | 4.6 | |
| AdaptiveSFTBackbone=Qwen3-8B2026.07 | 4.6 | |
| AdaptiveCLSBackbone=Qwen3-4B2026.07 | 4.3 | |
| AdaptiveCLSBackbone=Qwen3-8B2026.07 | 4.3 | |
| FixedThinkBackbone=Qwen3-4B2026.07 | 4.1 | |
| FixedThink+RLBackbone=Qwen3-4B2026.07 | 4.1 | |
| FixedNoThinkBackbone=Qwen3-8B2026.07 | 4.1 | |
| AdaptiveRLBackbone=Qwen3-4B2026.07 | 4 | |
| FixedNoThinkBackbone=Qwen3-4B2026.07 | 3.8 |