Tencent Hunyuan Hy4 launches: top open-source model gains vision and long-horizon agent skills
APPSO · wechat · 2026-11-04
Tencent's Hunyuan Hy4 official release upgrades agent capabilities and long-horizon tasks, topping multiple open-source benchmarks ahead of DeepSeek V4.1 Flash, GLM-5.3, and Kimi K3.1 Pro. Same parameters as the preview; gains come from significantly expanded pre- and post-training.
APPSo tested it in WorkBuddy:
- Vision: feed it screen recordings to clone game mechanics (built a mini MOBA in 30 min), turn room photos into 3D spaces, and iterate on its own outputs via continuous screenshotting
- Long-horizon work: /goal prompts keep targets intact through research-to-delivery game dev, with mid-task requirement changes handled gracefully
- Computer use: drove Blender to build a Singapore 'garden city' scene with 4,550 objects
- Office tasks: processed a folder of 10+ papers into 14 PPTs, one Excel, and a 19-page report in 30 minutes, with genuine reading comprehension
- Science: with a Peking University math team, found and formally verified a counterexample to the Méndez–Pinto–Villarreal conjecture
Pricing unchanged: ¥0.3/M cached input tokens, ¥6 uncached, ¥18 output; free in WorkBuddy Nov 4-15. Tencent is clearly binding model, tooling, and product into one iteration loop.
More from coding & agent
- 'Evals are fuzzy unit tests': Tejas Kumar's AI Engineer talk on evals gets peer recommendation — TejasKumar_ · 2026-11-04
- Document extraction past the POC: layout drift breaks pipelines, and parser model swaps add reprocessing pain — Mission-Leave-6551 · 2026-11-04
- SpecJudge author argues LLM-as-judge tools should never ask models for self-reported confidence — jokiruiz · 2026-11-04
- SOMA, the Claude Code CAD app, passes 1,000 users with a 9-minute bookshelf design walkthrough — Pr8tham · 2026-11-04
- How one AI agent cut serving costs 37% on 1M weekly messages, in 13 production changes — AccBalanced · 2026-11-04
- AI coding tool rex ships updates every 2 hours, echoing Cursor's early pace — sumitdotml · 2026-11-04