Selective Prediction on MMLU-Pro
94.5Sel-AUCStructU+Self-ConsU (TrueSkill+PageRank)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| StructU+Self-ConsU (TrueSkill+PageRank)Model=Claude 4.5 Sonnet, Scope=within2026.06 | 94.5 | 91 | |
| StructU+Self-ConsU (TrueSkill+PageRank)Model=DeepSeek R1, Scope=within2026.06 | 94.1 | 91.3 | |
| StructU+Self-ConsU (TrueSkill+PageRank)Model=GPT-OSS 20B, Scope=total2026.06 | 86.8 | 93.1 | |
| StructU+Self-ConsU (TrueSkill+PageRank)Model=Amazon Nova Premier, Scope=within2026.06 | 84.9 | 95.7 | |
| StructU+Self-ConsU (TrueSkill+PageRank)Model=Qwen 3 32B, Scope=within2026.06 | 82 | 97.2 |