VisBrowse-Bench (Full Toolset): 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; + IS setting: the authors' agent loop with text search, image search, reverse image search, image crop 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 | 47.6 | #60 |
| 2 | Kimi K2.5 | 41.4 | #139 |
| 3 | Gemini 3 Pro | 40.2 | #77 |
| 4 | Gemini 3 Flash | 39.1 | #93 |
| 5 | GPT-5.2 | 28.4 | #105 |
| 6 | Gemini 2.5 Pro | 26.6 | #145 |
| 7 | GPT-5.1 | 23.1 | #131 |
| 8 | Gemini 2.5 Flash | 20.7 | #237 |
| 9 | Claude Sonnet 4.6 | 18.3 | #85 |
| 10 | Qwen 3 VL 235B A22B Instruct | 14.2 | #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-full-toolset · How It Works · Data refreshed daily, snapshot 2026-10-11.