Tencent launches flagship Hy4 with native vision, plans open weights at same API price
智东西 · wechat · 2026-11-04
Tencent released Hy4, the final version of its flagship model, with plans to open-source the weights. It's live on Tencent Cloud and OpenRouter at the same API price as Hy4 Preview.
- Upgrades: Much larger pre/post-training than August's Hy4 Preview, first-ever native vision, and major gains in instruction following, long-context, long-horizon tasks and deep research; beats DeepSeek-V4.1-Flash on benchmarks but still trails GPT-6.1 Sol and Claude Opus 5.5.
- Hands-on highlights (Jiqizhixin preview): screenshot-driven code editing, image-to-3D modeling in Blender (first version in 17 minutes, clearly improved over 6 feedback rounds), auto-generated a SpaceX Q3 earnings deck, GUI control of WeChat for posting comments and summarizing group chats; held its goal across a 500K-token 3D task.
- RSI: With research agent Hyra 2.0, it optimized training infrastructure for smaller same-architecture models, boosting end-to-end throughput by 20.1% and auto-training a Jev model that beats the official one; open-sourced HyCell (16M params matching SOTA) and HyPDE.
- Safety: Offensive cybersecurity data removed, defensive-only retained; bio experiment-design capability remains well below frontier models.
- Ecosystem: vLLM/SGLang support plus an Ascend 950DT deployment guide landed pre-launch; a cheaper, faster Hy4 Flash is coming soon.
More from Models
- Hobbyist rewrites Qwen 0.6B weights with 39.8M tokens, no gradients, in 3h12m — Altruistic_Heat_9531 · 2026-11-04
- Same leak, reshared: ChatGPT's five family plan prices spotted again — heyshrutimishra · 2026-11-04
- KOL visits StepFun's Shanghai office, hints a powerful new model is coming — teortaxesTex · 2026-11-04
- Fable 5.5 delayed as Anthropic reportedly bets OpenAI's Astra 6.2 won't beat it — haider1 · 2026-11-04
- ByteDance open-sources Axion 1.0-35B agent foundation model with full reproducible recipe — pmttyji · 2026-11-04
- Community asks Moonshot when Kimi K3.1 model weights will drop — bgmshana · 2026-11-04