Open-Ended Question Answering on SAGE Web Search
35.1Weighted Recall (Com. Sci.)GPT-5
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-5Avg. Searches=13.69, Avg. Refs=29.022026.02 | 35.1 | 25 | 18.8 | 26.2 | 26.3 | |
| gte-Qwenk=10, Avg. Searches=4.54, Avg. Refs=29.32026.02 | 28.9 | 33.5 | 36.6 | 33 | 33 | |
| Gemini-2.5-flashAvg. Searches=6.50, Avg. Refs=22.122026.02 | 28.2 | 10.1 | 15.7 | 9.8 | 16 | |
| GPT-5-miniAvg. Searches=10.07, Avg. Refs=30.702026.02 | 27.4 | 17.8 | 13.9 | 22.4 | 20.4 | |
| GPT-5-nanoAvg. Searches=8.59, Avg. Refs=30.272026.02 | 25.9 | 29.8 | 15.7 | 21.1 | 20.6 | |
| gte-Qwenk=5, Avg. Searches=4.79, Avg. Refs=15.42026.02 | 22.4 | 26.3 | 29.3 | 26.2 | 26 | |
| BM25k=10, Avg. Searches=4.17, Avg. Refs=29.92026.02 | 20.3 | 36.2 | 34 | 32.3 | 30.7 | |
| Gemini-2.5-proAvg. Searches=10.77, Avg. Refs=14.182026.02 | 20 | 5.2 | 12.2 | 6.7 | 11 | |
| BM25k=5, Avg. Searches=4.72, Avg. Refs=16.72026.02 | 18.2 | 29.4 | 28.3 | 26.3 | 25.5 | |
| DR TuluAvg. Searches=4.25, Avg. Refs=35.952026.02 | 18 | 17.2 | 14.1 | 20.2 | 17.4 | |
| ReasonIRk=10, Avg. Searches=4.44, Avg. Refs=26.82026.02 | 16.1 | 29.5 | 32 | 27.4 | 26.2 | |
| ReasonIRk=5, Avg. Searches=6.21, Avg. Refs=15.82026.02 | 11.3 | 18.5 | 22.3 | 17 | 17.3 |