VisBrowse-Bench (Text Search): leaderboard
Metric: Accuracy (%) on VisBrowse-Bench's 169 expert-built visual multi-hop search questions (7 domains), final answer extracted by regular expression and judged against the reference by GPT-5.1; + TS setting: the authors' agent loop with text search and webpage visit tools; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 42.6 | #60 |
| 2 | Gemini 3 Pro | 38.5 | #77 |
| 3 | Gemini 3 Flash | 37.9 | #93 |
| 4 | GPT-5.2 | 26 | #105 |
| 5 | Claude Sonnet 4.6 | 23.7 | #85 |
| 6 | Kimi K2.5 | 21.9 | #139 |
| 7 | Gemini 2.5 Pro | 20.7 | #145 |
| 8 | Gemini 2.5 Flash | 17.2 | #237 |
| 9 | GPT-5.1 | 16.6 | #131 |
| 10 | Qwen 3 VL 235B A22B Instruct | 11.2 | #264 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=visbrowse-bench-text-search · How It Works · Data refreshed daily, snapshot 2026-10-11.