K-BrowseComp: leaderboard
K-BrowseComp (2026) is a web-browsing agent benchmark grounded in Korean contexts, with human-written multi-hop and parallel-constraint questions validated by native speakers. This board is Pass@1 accuracy in percent on the 300-question verified subset, for a deep-research agent limited to 10 Perplexity searches per question; higher is better.
Metric: Pass@1 accuracy (%) on the 300-question K-BrowseComp-Verified subset; search_evals deep-research agent with Perplexity Search, 10 search calls per question, one run, GPT-5.4-mini answer extraction against the gold answer. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 Mini | 30.67 | #389 |
| 2 | GLM-5.1 | 30.67 | #179 |
| 3 | DeepSeek V4 Pro | 30 | #188 |
| 4 | Gemma 4 31B (IT) | 23.33 | #342 |
| 5 | Qwen 3.6 35B A3B | 12 | #462 |
| 6 | Gemini 3.1 Flash Lite | 11.33 | #364 |
| 7 | K-EXAONE | 10.33 | #708 |
| 8 | A.X 4.0 | 5.33 | #863 |
| 9 | HyperCLOVA X SEED Think (32B) | 2.33 | #1437 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=k-browsecomp · How It Works · Data refreshed daily, snapshot 2026-11-13.