Abstention on AbstentionBench UMWP
80.9Abstention F1GPT-5.2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.2Model Type=Proprietary API Model2026.05 | 80.9 | 67.9 | 100 | 97.9 | |
| Claude Sonnet 4.5Model Type=Proprietary API Model2026.05 | 76.7 | 62.3 | 100 | 100 | |
| Gemini 3Model Type=Proprietary API Model2026.05 | 76.7 | 62.3 | 100 | 95.7 | |
| TIARBackbone=Llama-3.1-8B-Instruct2026.05 | 72.3 | 56.6 | 100 | 95.7 |