EvoBrowseComp-EN: leaderboard
EvoBrowseComp (2026) tests web-browsing agents on 400 English multi-hop questions built from knowledge that appeared on the live web after model training cutoffs, so answers need genuine search rather than recall. Scored by accuracy with search and visit tools, judged by an LLM against the reference; higher is better.
Metric: Accuracy (%) on the 400 English EvoBrowseComp questions (complex multi-hop questions synthesized from fresh live-web knowledge to resist parametric recall), search agent with web search and page visit tools (Tongyi DeepResearch tool definitions), answers judged correct against the reference by a GLM-5-Chat judge; temperature 0.6, top_p 0.95, 128K context, at most 40 tool calls; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 (Max) | 46.2 | #56 (Claude Opus 4.8) |
| 2 | Claude Opus 4.6 (Max) | 44.8 | #102 (Claude Opus 4.6) |
| 3 | Qwen 3.5 397B A17B | 42 | #246 |
| 4 | GLM-5 | 39.2 | #260 |
| 5 | DeepSeek V4 Flash (High) | 34.5 | #258 (DeepSeek V4 Flash) |
| 6 | Kimi K2.6 | 31 | #178 |
| 7 | Qwen 3.5 35B A3B | 29.2 | #482 |
| 8 | Qwen 3.5 122B A10B | 29.2 | #331 |
| 9 | DeepSeek V4 Flash (Non-reasoning) | 27.2 | #258 (DeepSeek V4 Flash) |
| 10 | Qwen 3.5 27B | 25 | #359 |
| 11 | DeepSeek V3.2 (Thinking) | 23 | #406 (DeepSeek V3.2) |
| 12 | Qwen 3 235B A22B 2507 (Thinking) | 16.5 | #463 |
| 13 | DeepSeek V4 Pro (Max) | 16.5 | #188 (DeepSeek V4 Pro) |
| 14 | DeepSeek V4 Flash (Max) | 16.5 | #258 (DeepSeek V4 Flash) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=evobrowsecomp-en · How It Works · Data refreshed daily, snapshot 2026-11-13.