all agent alignment api update architecture benchmark business cuda dataset deployment embedding evaluation fine tuning game dev hype inference infrastructure library ner new model open source optimization plugin probe targeted prompt engineering quantization rag research security testing tool training tutorial workflow
VibeSearchBench — 🔍 The hardest search benchmark in the wild — vague, multi-turn, proactive. 200 long-horizon tasks with persona-driven progressive disclosure, scored by verifiable schema-free knowledge-graph evaluation. No vibes, just triplet F1.
GitHub Trending AI · 13d ago · 7 · benchmark agent open source evaluation

VibeSearchBench is a new benchmark for evaluating multi-turn agentic search systems with 200 tasks involving vague queries and progressive user disclosure, using knowledge-graph-based evaluation metrics (precision/recall/F1 at node and triplet levels). The benchmark integrates with OpenAI-compatible LLMs and OpenClaw CLI, making it directly applicable for engineers building and testing agentic search workflows.