LLM Stats (BrowseComp): leaderboard
BrowseComp poses 1,266 questions with short answers that are hard to find but easy to check; writers confirmed that five plain searches did not surface the answer. The agent browses the live web, and a grader model judges whether its final answer matches the reference. LLM Stats lists these figures as self-reported; higher is better. Source: arxiv.org/abs/2504.12516.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 68 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 91.5 | #6 |
| 2 | Kimi K3 | 91.2 | #30 |
| 3 | Claude Opus 5 | 90.8 | #12 |
| 4 | GPT-5.6 Sol | 90.4 | #19 |
| 5 | GPT-5.5 Pro | 90.1 | #17 |
| 6 | GPT-5.6 Terra | 87.5 | #41 |
| 7 | Claude Mythos Preview | 86.9 | #4 |
| 8 | Kimi K2.6 | 86.3 | #104 |
| 9 | Seed 2.1 Pro | 86.2 | #67 |
| 10 | Gemini 3.1 Pro (Preview) | 85.9 | #58 |
| 11 | Seed 2.1 Turbo | 84.9 | #86 |
| 12 | Claude Sonnet 5 | 84.7 | #57 |
| 13 | GPT-5.5 | 84.4 | #31 |
| 14 | Claude Opus 4.8 | 84.3 | #37 |
| 15 | Hy3 | 84.2 | #93 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-browsecomp · How It Works · Data refreshed daily, snapshot 2026-10-20.