Question Answering on MIP and AbstentionBench
100Answer ScoreBase
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaseBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 100 | 91.8 | 1,687.3 | 0.544 | |
| BaseBackbone=Qwen3-30B-A3B-Thinking2026.05 | 100 | 97.5 | 2,658.5 | 0.367 | |
| PromptingBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 99.6 | 89.1 | 1,046.5 | 0.851 | |
| Plain RLBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 99.4 | 90.3 | 1,769.7 | 0.51 | |
| Plain RLBackbone=Qwen3-30B-A3B-Thinking2026.05 | 99 | 96.1 | 1,694.1 | 0.567 | |
| PromptingBackbone=Qwen3-30B-A3B-Thinking2026.05 | 98.8 | 94 | 1,440.1 | 0.653 | |
| JTSBackbone=Qwen3-30B-A3B-Thinking2026.05 | 94.7 | 93.4 | 810.4 | 1.152 | |
| JTSBackbone=DeepSeek-R1-Distill-Qwen-14B2026.05 | 92.4 | 86.8 | 844.7 | 1.028 |