Abstention on AbstentionBench QAQA
75.8Abstention F1Gemini 3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini 3Model Type=Proprietary API Model2026.05 | 75.8 | 66.7 | 87.8 | 78.3 | |
| Claude Sonnet 4.5Model Type=Proprietary API Model2026.05 | 66.7 | 55.6 | 83.3 | 56.5 | |
| GPT-5.2Model Type=Proprietary API Model2026.05 | 66 | 59.3 | 74.4 | 65.2 | |
| TIARBackbone=Llama-3.1-8B-Instruct2026.05 | 65.9 | 51.9 | 90.3 | 43.5 |