Abstention on AbstentionBench BB Known Unk.
95.8Abstention F1TIAR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TIARBackbone=Llama-3.1-8B-Instruct2026.05 | 95.8 | 100 | 92 | 82.6 | |
| Claude Sonnet 4.5Model Type=Proprietary API Model2026.05 | 93.9 | 100 | 88.5 | 82.6 | |
| GPT-5.2Model Type=Proprietary API Model2026.05 | 93.6 | 95.7 | 91.7 | 82.6 | |
| Gemini 3Model Type=Proprietary API Model2026.05 | 84.4 | 82.6 | 86.4 | 82.6 |