BrowseComp Long Context - 128K
General QA · 2025-08-07
BrowseComp Long Context gives the model a BrowseComp-style question together with a long list of relevant search results and asks it to answer from that context alone, without browsing. This row is the 128K-input bin. Scores are answer accuracy. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.2 | 92.0 |
| GPT-5.1 | 90.0 |
| GPT-5 | 90.0 |
| GPT-5 Mini | 89.4 |
| GPT-4.1 Nano | 89.4 |
| GPT-4.1 Mini | 89.0 |
| o3 | 88.3 |
| GPT-4.1 | 85.9 |
| GPT-5 Nano | 80.4 |
| O4 Mini | 80.0 |