Question Answering on QASC (Leakage Error, Answerability Accuracy)
14Leakage ErrorSeedRG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SeedRGModel=GPT-52026.05 | 14 | 71 | |
| SeedRGModel=Gemini 2.5 Flash2026.05 | 16 | 76 | |
| SeedRGModel=Claude Sonnet 4.52026.05 | 17 | 69 | |
| Direct Generation (DG)Model=GPT-52026.05 | 32 | 41 | |
| Direct Generation (DG)Model=Gemini 2.5 Flash2026.05 | 38 | 38 | |
| Direct Generation (DG)Model=Claude Sonnet 4.52026.05 | 53 | 29 | |
| Original BenchmarkModel=GPT-52026.05 | 75 | 18 | |
| Original BenchmarkModel=Gemini 2.5 Flash2026.05 | 77 | 21 | |
| Original BenchmarkModel=Claude Sonnet 4.52026.05 | 82 | 17 |