Jev releases 1M public model decisions across 7 datasets, reading 2B tokens
MaziyarPanahi · x · 2026-11-09
Maziyar Panahi announced the Jev datasets: 1 million model decisions now fully public, generated by reading 2 billion tokens across 7 subsets — agent tool decisions (180K), agent trace audits (100K), policy verification (180K), safety and refusal (150K), intent/domain routing (130K), customer support (130K), and RAG claim checks (130K).
Every row includes a source label and the model's answer, with a single split ensuring no group appears in both train and test. The author suggests training a small judge model on it for internal quality control, e.g. in hospital settings.
Related event: Developer Open-Sources 1M Model Decision Dataset, Trained for Just $5(2 posts)→
More from coding & agent
- HF's Niels Rogge releases 'Getting Started with Local AI' slides: quantization, MoE, local agents — NielsRogge · 2026-11-09
- OpenClaw runs "massively multiplayer online AI": hundreds of agents on one server — vincent_koc · 2026-11-09
- Loop: a mobile-first feed-based orchestrator for your coding agents — Two_Half_Pints · 2026-11-09
- Qwen Code Desktop ships v0.27.1 with telemetry and session fixes — github-actions[bot] · 2026-11-09
- dair-ai founder: own your evals — public benchmarks won't tell you what to ship — dair_ai · 2026-11-09
- office-engine handles 500M CSV rows in a browser tab at ~1 GB RAM — pierreneter · 2026-11-09