Hy4, top open-source agent, builds a 6-slide deck from one prompt and 40 sources
aftahi_ai · x · 2026-11-07
A hands-on test of Hy4 inside WorkBuddy: one prompt to research the open-source AI model landscape and produce a 6-slide deck. It pulled 40 sources, built charts and comparison tables, and even caught and fixed a layout issue on its own slide. Notably, it flagged that some benchmark scores aren't directly comparable across versions instead of just presenting numbers. Context: Hy4 is currently #1 open-source on Agent Arena, priced around $0.78 per task per Artificial Analysis; Tencent scaled WorkBuddy capacity 10x after launch demand.
More from Models
- RoboColosseum launches as open real-world robot eval platform with 100+ robots — DJiafei · 2026-11-08
- Two Minute Papers Explains the OpenAI Astra Weight-Space Programming Hack — Two Minute Papers · 2026-11-08
- Don't trust LLM 'emergence' claims — read 'emergence' as 'training', researcher tells students — tallinzen · 2026-11-08
- MaxxBench: models clustered at 74% on DeepSWE spread 23 points when tasks change slightly — dyn___ · 2026-11-08
- Gemini Live voice glitch allegedly cloned the user's own voice mid-call — iruletheworldmo · 2026-11-08
- Two ~500B MoE Post-Training Designs Compared: Separate RL Runs vs Joint RL — himanshustwts · 2026-11-08