VibeSearchBench: leaderboard
Long-horizon proactive web search: across 200 expert-written professional and everyday tasks (half Chinese, half English), an agent works with a simulated user who reveals needs one stage at a time, then extracts a knowledge graph of what it found; VibeSearchBench (Xiaohongshu, 2026). Triplet F1 in percent against the annotated ground-truth graph, judged by an LLM and averaged over the Pro and Daily subsets and 3 runs, under the OpenClaw agent harness; higher is better.
Metric: Triplet F1 (%; an LLM judge matches the knowledge graph the agent extracts after a multi-turn session with a progressive-disclosure user simulator against the annotated ground-truth graph; OpenClaw agent harness; mean of VibeSearch-Pro and VibeSearch-Daily; mean of 3 runs per task). Source: arxiv.org. Saturation forecast: Around April 2028. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 30.3 | #100 |
| 2 | DeepSeek V4 Pro | 27.64 | #181 |
| 3 | Kimi K2.6 | 26.17 | #173 |
| 4 | Seed 2.0 Pro | 25.23 | #209 |
| 5 | Gemini 3.1 Pro (Preview) | 23.62 | #91 |
| 6 | Qwen 3.5 397B A17B | 23.1 | #242 |
| 7 | GPT-5.4 | 21.92 | #121 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vibesearchbench · How It Works · Data refreshed daily, snapshot 2026-11-12.