Tencent Open-Sources Hunyuan Hy3 for Coding and Agents

智东西 · wechat · 2026-07-06

Tencent has officially released and open-sourced the Hunyuan Hy3 model. Featuring an MoE architecture, it has 295B total parameters and 21B activated parameters, supporting up to a 256K context length and targeting Coding and Agent scenarios. Compared to April's preview version, its agent and coding capabilities have improved by 20-30%. On benchmarks, it scored 57.9 on SWE-bench Pro, surpassing DeepSeek V4pro and Seed2.1 Pro, and ranked first on FrontierScience-Olympiad, beating GPT5.5 and Qwen3.7 Max. Pricing is set at 1 yuan/million tokens for input and 4 yuan/million tokens for output. Tencent Yuanbao has immediately integrated the model and launched Agent capabilities, enabling the direct delivery of usable files within chats.

Related event: Tencent Open-Sources Hunyuan Hy3 for Agentic Workloads(26 posts)→

Original post →

More from Models

Models channel →