Abstention on AbstentionBench KUQ Cont
100Abstention F1Gemini 3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini 3Model Type=Proprietary API Model2026.05 | 100 | 100 | 100 | 50 | |
| GPT-5.2Model Type=Proprietary API Model2026.05 | 92.9 | 92.9 | 92.9 | 75 | |
| TIARBackbone=Llama-3.1-8B-Instruct2026.05 | 92.9 | 92.9 | 92.9 | 75 | |
| Claude Sonnet 4.5Model Type=Proprietary API Model2026.05 | 92.3 | 85.7 | 100 | 100 |