Abstention on AbstentionBench FreshQA
75.2Abstention F1Claude Sonnet 4.5
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Claude Sonnet 4.5Model Type=Proprietary API Model2026.05 | 75.2 | 73.2 | 77.4 | 63.6 | |
| TIARBackbone=Llama-3.1-8B-Instruct2026.05 | 68.4 | 71.4 | 65.6 | 52.3 | |
| GPT-5.2Model Type=Proprietary API Model2026.05 | 48.8 | 35.7 | 76.9 | 70.5 | |
| Gemini 3Model Type=Proprietary API Model2026.05 | 32.8 | 19.6 | 100 | 81.8 |