Tencent Releases Open-Source MoE Model Hunyuan Hy3
eyishazyer · x · 2026-07-09
Tencent has officially released the Hunyuan Hy3 large language model, marking it as one of the most notable open-source releases of the year.
The model utilizes a 295B parameter MoE architecture with 21B activated parameters and supports a 256K context length. It delivers outstanding performance in coding and reasoning, is specifically designed for agentic workflows, and is open-sourced under the Apache 2.0 license.
Related event: Tencent Releases Open-Source Hunyuan Hy3 Model for Coding and Agents(9 posts)→
More from Models
- Google is reportedly pre-training Gemini 4 as a completely new foundation model — kimmonismus · 2026-07-21
- Google has reportedly started pre-training Gemini 4 as a new foundation model — kimmonismus · 2026-07-21
- Google says its most ambitious Gemini 4 pre-training run has started — OfficialLoganK · 2026-07-21
- Google launches Gemini 3.6 Flash, Flash-Lite, and Flash Cyber with new pricing — fofrAI · 2026-07-21
- GPT’s “dream” about a Jacobian-conjecture proof becomes a perfect AI meme — BLUECOW009 · 2026-07-21
- Fable 5 reaches 34% pass@1 on FrogsGame after 17 hours and 25M tokens — aronchick · 2026-07-21