VibeSearchBench (ReAct) - Pro: leaderboard

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; ReAct loop with self-summarising context compaction; 100 professional-domain VibeSearch-Pro tasks; mean of 3 runs per task). Source: arxiv.org. Saturation forecast: Around 2029. 7 models tracked.

Top models

#ModelScore
1Claude Opus 4.629.79
2DeepSeek V4 Pro28.7
3Kimi K2.627.12
4Seed 2.0 Pro25.86
5Qwen 3.5 397B A17B23.65
6Gemini 3.1 Pro (Preview)22.41
7GPT-5.418.94

Interactive version: theaggregate.ai/benchmark?slug=vibesearchbench-react-pro · How It Works · Data refreshed daily, snapshot 2026-09-26.