BrowseComp-V3: leaderboard
Metric: Success rate (%): share of questions answered correctly on 300 multimodal multi-hop browsing questions (Science 50, Technology 50, Society 75, Culture 75, Life 50) whose evidence is split across text and images on public web pages; OmniSeeker, the paper's browsing agent with TextSearch (Serper, top 5), WebVisit (Jina), ImageSearch, ImageCrop and ReverseImageSearch tools, up to 20 rounds per question, pass@1; higher is better. Source: arxiv.org. Saturation forecast: Around March 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 36 | #105 |
| 2 | Seed 1.8 | 33.67 | #136 |
| 3 | O4 Mini | 26 | #172 |
| 4 | Gemini 3 Flash (Preview) | 23.67 | #78 |
| 5 | Claude Sonnet 4.5 | 22.67 | #138 |
| 6 | MiMo-V2-Flash | 16.67 | #359 |
| 7 | Qwen 3 VL 235B A22B Instruct | 14.33 | #264 |
| 8 | GPT-4o | 11.41 | #333 |
| 9 | Qwen 3 VL 8B Instruct | 5.33 | #401 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=browsecomp-v3 · How It Works · Data refreshed daily, snapshot 2026-10-11.