VisBrowse-Bench (Direct Answer): 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; Direct Answer setting: no tools are provided, the model answers from its own knowledge; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 32.5 | #93 |
| 2 | Claude Opus 4.6 | 27.2 | #60 |
| 3 | Gemini 3 Pro | 23.7 | #77 |
| 4 | Kimi K2.5 | 21.3 | #139 |
| 5 | Gemini 2.5 Pro | 19.5 | #145 |
| 6 | GPT-5.2 | 14.8 | #105 |
| 7 | Claude Sonnet 4.6 | 13 | #85 |
| 8 | GPT-5.1 | 13 | #131 |
| 9 | Qwen 3 VL 235B A22B Instruct | 10.1 | #264 |
| 10 | Gemini 2.5 Flash | 9.5 | #237 |
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-direct-answer · How It Works · Data refreshed daily, snapshot 2026-10-11.