Question Answering on HotpotQA (Leakage Error, Answerability Accuracy)
1.4Leakage ErrorSeedRG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SeedRGModel=GPT-52026.05 | 1.4 | 41.8 | |
| SeedRGModel=Gemini 2.5 Flash2026.05 | 4.1 | 47.3 | |
| SeedRGModel=Claude Sonnet 4.52026.05 | 13.5 | 59.5 | |
| Direct Generation (DG)Model=Gemini 2.5 Flash2026.05 | 18 | 46 | |
| Direct Generation (DG)Model=GPT-52026.05 | 19 | 50 | |
| Direct Generation (DG)Model=Claude Sonnet 4.52026.05 | 31 | 47 | |
| Original BenchmarkModel=Gemini 2.5 Flash2026.05 | 31.1 | 50 | |
| Original BenchmarkModel=Claude Sonnet 4.52026.05 | 45.9 | 47.3 | |
| Original BenchmarkModel=GPT-52026.05 | 50 | 33.8 |