Question Answering on NQ-Open
47.4Exact Match (EM)MD
Evaluation Results
| Method | Links | |
|---|---|---|
| MDModel=Llama 32025.03 | 47.4 | |
| PTRModel=GPT-5.42026.04 | 46.9 | |
| MARGOBackbone=Qwen3-8B2026.07 | 42.24 | |
| FixedThink+RLBackbone=Qwen3-8B2026.07 | 40.66 | |
| MDModel=Llama 22025.03 | 40.5 | |
| AdaptiveRLBackbone=Qwen3-8B2026.07 | 40.08 | |
| FixedThinkBackbone=Qwen3-8B2026.07 | 39.94 | |
| GreedyModel=Llama 32025.03 | 39.6 | |
| FixedNoThink+RLBackbone=Qwen3-8B2026.07 | 39.09 | |
| FSCModel=Llama 32025.03 | 38.7 | |
| FixedNoThinkBackbone=Qwen3-8B2026.07 | 38.64 | |
| USCModel=Llama 22025.03 | 38.6 | |
| BoNModel=Llama 32025.03 | 38.6 | |
| IDModel=Llama 32025.03 | 38.5 | |
| AdaptiveCLSBackbone=Qwen3-8B2026.07 | 38.5 | |
| AdaptiveSFTBackbone=Qwen3-8B2026.07 | 38.5 | |
| PTRModel=Claude Sonnet 4.62026.04 | 38 | |
| SRModel=Llama 22025.03 | 37.3 | |
| USCModel=Llama 32025.03 | 37 | |
| SRModel=Llama 32025.03 | 36.9 | |
| GreedyModel=Llama 22025.03 | 36.6 | |
| DoLaModel=Llama 32025.03 | 35.7 | |
| BoNModel=Llama 22025.03 | 35.2 | |
| MARGOBackbone=Qwen3-4B2026.07 | 34.76 | |
| FixedThink+RLBackbone=Qwen3-4B2026.07 | 33.8 | |
| DoLaModel=Llama 22025.03 | 33.7 | |
| FSCModel=Llama 22025.03 | 33.5 | |
| IDModel=Llama 22025.03 | 33.1 | |
| FixedThinkBackbone=Qwen3-4B2026.07 | 32.94 | |
| AdaptiveRLBackbone=Qwen3-4B2026.07 | 32.49 | |
| MDModel=Gemma 22025.03 | 31 | |
| FixedNoThink+RLBackbone=Qwen3-4B2026.07 | 30.61 | |
| AdaptiveCLSBackbone=Qwen3-4B2026.07 | 30.33 | |
| AdaptiveSFTBackbone=Qwen3-4B2026.07 | 30.28 | |
| PTRModel=Claude Haiku 4.52026.04 | 30 | |
| FixedNoThinkBackbone=Qwen3-4B2026.07 | 29.5 | |
| SRModel=Gemma 22025.03 | 28 | |
| PTRModel=GPT-4o-Mini2026.04 | 26 | |
| FSCModel=Gemma 22025.03 | 24 | |
| USCModel=Gemma 22025.03 | 23.9 | |
| GreedyModel=Gemma 22025.03 | 23 | |
| BoNModel=Gemma 22025.03 | 19 | |
| DoLaModel=Gemma 22025.03 | 18.2 | |
| IDModel=Gemma 22025.03 | 16.5 | |
| ReActModel=GPT-5.42026.04 | 16 | |
| ReActModel=GPT-4o-Mini2026.04 | 6 | |
| ReActModel=Claude Haiku 4.52026.04 | 2 | |
| ReActModel=Claude Sonnet 4.62026.04 | 0 |