Tencent Open-Sources Hunyuan Hy4 Preview: 770B Parameters, 1M Context

Tencent Hunyuan released and open-sourced its next-generation flagship model, Hy4 Preview, on August 27–28. Built on a MoE architecture with 770B total parameters and 49B active, it offers a 1 million token context window and is fully open-sourced under Apache 2.0 on HuggingFace and GitHub, along with a blog post and an arXiv paper. Tencent positions it as an open-source frontier model for productivity (coding, office work, game development, scientific research), with API pricing of 6 yuan per million tokens for input, 18 yuan for output, and 0.3 yuan for cached input.

Confirmed

• Scale: 770B total parameters, 49B active (A49B), doubling Hy3; supports 1M context; open-sourced under Apache 2.0, with the model available on HuggingFace (tencent/Hy4-preview) and GitHub, plus an arXiv paper
• Pricing: 6 yuan per million tokens for input, 18 yuan for output, and 0.3 yuan for cache
• Official evaluations put it roughly on par with GLM-5.3, with slightly better internal blind-test scores than GLM-5.3 and the KimiK series
• Hands-on reports: APPSO says its CodingAgent long-horizon execution ranks in the first tier, capable of independently building a 3D lunar rover game with a physics engine and fixing bugs in a travel expense system; Cyber Chanxin demonstrated game development skills by connecting to the Unreal 5 engine via MCP

Unconfirmed

• @xeophon's relayed claim that its "benchmark scores match GLM-5.3" comes mainly from officially released data and still lacks broad third-party independent verification

Why it matters

• A fully open 770B-class model with 1M context and low pricing further lowers the barrier to frontier capabilities, directly shaking up the open-source community and China's LLM competitive landscape (against GLM, Kimi, etc.); its performance in CodingAgent and game engine integration showcases the potential of large models in real production workflows.

2026-08-27 ~ 2026-08-28 · 7 related posts

• Tencent Releases Hy4-Preview: New MoE-Based Conversational Model — tencent · 2026-08-27
• [source] Tencent releases Hy4preview: 770B open-source model with 1M context — 腾讯混元 · 2026-08-28
• Tencent open-sources Hy4 Preview: 770B params, 1M context, Apache 2.0 — xeophon · 2026-08-28
• Tencent open-sources Hy4 Preview: 770B params, 1M context, on par with GLM-5.3 — xeophon · 2026-08-28
• [source] Tencent Hunyuan Releases Hy4 Preview: 770B Params, 1M Context — TencentHunyuan · 2026-08-28
• [source] Tencent releases Hy4preview: 770B params, 1M context, and Unreal Engine game generation — 赛博禅心 · 2026-08-28
• Tencent Releases Hunyuan Hy4: 770B Params, CodingAgent Enters First Tier — APPSO · 2026-08-28