Mathematical Reasoning on Math Benchmarks
76.9Accuracy (%)R1-Distill-Qwen3-8B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| R1-Distill-Qwen3-8BContext Length=2048, Monitoring Strategy=None2025.09 | 76.9 | 0 | 2,048 | 100 | |
| R1-Distill-Qwen3-8BContext Length=4096, Monitoring Strategy=None2025.09 | 76.9 | 0 | 4,096 | 100 | |
| BoostAbstentionBase Model=R1-Distill-Qwen3-8B, Context Length=20482025.09 | 76.9 | 0 | 2,048 | 100 | |
| BoostAbstentionBase Model=R1-Distill-Qwen3-8B, Context Length=40962025.09 | 76.9 | 0 | 4,096 | 100 | |
| GPT-oss-20BContext Length=2048, Monitoring Strategy=None2025.09 | 74.6 | 0 | 2,029 | 97 | |
| GPT-oss-20BContext Length=4096, Monitoring Strategy=None2025.09 | 74.6 | 0 | 4,015 | 97 | |
| BoostAbstentionBase Model=GPT-oss-20B, Context Length=20482025.09 | 74.6 | 0 | 2,023 | 97 | |
| BoostAbstentionBase Model=GPT-oss-20B, Context Length=40962025.09 | 74.6 | 0 | 3,933 | 84.8 | |
| MonitorexpressBase Model=GPT-oss-20B, Context Length=20482025.09 | 74.6 | 100 | 1,135 | 21.2 | |
| MonitorexpressBase Model=GPT-oss-20B, Context Length=40962025.09 | 74.6 | 100 | 1,499 | 15.2 | |
| MonitorexpressBase Model=R1-Distill-Qwen3-8B, Context Length=20482025.09 | 74.6 | 63.3 | 1,455 | 40 | |
| MonitorexpressBase Model=R1-Distill-Qwen3-8B, Context Length=40962025.09 | 74.6 | 63.3 | 2,551 | 30 | |
| MonitorhiddenBase Model=R1-Distill-Qwen3-8B, Context Length=20482025.09 | 73.8 | 97.9 | 787 | 22.2 | |
| MonitorhiddenBase Model=R1-Distill-Qwen3-8B, Context Length=40962025.09 | 73.8 | 97.9 | 1,243 | 0 | |
| MonitorhiddenBase Model=GPT-oss-20B, Context Length=20482025.09 | 73.7 | 98.9 | 1,784 | 78.4 | |
| MonitorhiddenBase Model=GPT-oss-20B, Context Length=40962025.09 | 73.7 | 98.9 | 3,317 | 5.4 | |
| QwQ-32BContext Length=2048, Monitoring Strategy=None2025.09 | 71.5 | 0 | 2,048 | 100 | |
| QwQ-32BContext Length=4096, Monitoring Strategy=None2025.09 | 71.5 | 0 | 4,096 | 100 | |
| BoostAbstentionBase Model=QwQ-32B, Context Length=20482025.09 | 71.5 | 0 | 2,048 | 100 | |
| BoostAbstentionBase Model=QwQ-32B, Context Length=40962025.09 | 71.5 | 0 | 4,096 | 100 | |
| MonitorexpressBase Model=QwQ-32B, Context Length=20482025.09 | 71.5 | 100 | 1,164 | 32.4 | |
| MonitorexpressBase Model=QwQ-32B, Context Length=40962025.09 | 71.5 | 100 | 1,828 | 32.4 | |
| MonitorhiddenBase Model=QwQ-32B, Context Length=20482025.09 | 71.5 | 96.9 | 262 | 0 | |
| MonitorhiddenBase Model=QwQ-32B, Context Length=40962025.09 | 71.5 | 96.9 | 262 | 0 | |
| R1-Distill-Qwen-32BContext Length=2048, Monitoring Strategy=None2025.09 | 56.2 | 0 | 2,048 | 100 | |
| R1-Distill-Qwen-32BContext Length=4095, Monitoring Strategy=None2025.09 | 56.2 | 0 | 4,095 | 97.4 | |
| BoostAbstentionBase Model=R1-Distill-Qwen-32B, Context Length=20482025.09 | 56.2 | 0 | 2,005 | 94.7 | |
| BoostAbstentionBase Model=R1-Distill-Qwen-32B, Context Length=40962025.09 | 56.2 | 0 | 3,903 | 91.2 | |
| MonitorexpressBase Model=R1-Distill-Qwen-32B, Context Length=20482025.09 | 56.2 | 100 | 801 | 0 | |
| MonitorexpressBase Model=R1-Distill-Qwen-32B, Context Length=40962025.09 | 56.2 | 100 | 801 | 0 | |
| MonitorhiddenBase Model=R1-Distill-Qwen-32B, Context Length=20482025.09 | 55.3 | 97 | 457 | 5.3 | |
| MonitorhiddenBase Model=R1-Distill-Qwen-32B, Context Length=40962025.09 | 55.3 | 97 | 565 | 5.3 |