FrontierThe story, in brief
Tencent Releases Hy3: An Open 295B Mixture-of-Experts (MoE) Model with 21B Active Parameters and 256K Context
295B parameters. 21B active. Tencent's Hy3 just hit 78.0 on SWE-Bench Verified—and it's open-source.
The KeyNews take
Why it matters
Tencent enters the competitive open-model landscape with a efficient MoE architecture that challenges closed-model reasoning performance. The Apache 2.0 release and free trial signal aggressive positioning in the post-GPT-5 model wars.
The key facts
7 to knowHy3: 295B total parameters, 21B active per token (MoE)
256K context window
78.0 on SWE-Bench Verified
Apache 2.0 open-source license
Lower hallucination rates reported
Free access via OpenRouter through July 21, 2026
Targets reasoning, agentic, and long-context tasks
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Tencent's Hy team released Hy3, a 295B Mixture-of-Experts (MoE) model that activates only 21B parameters per token. It ships under Apache 2.0 with a 256K context window, targeting reasoning, agentic, and long-context tasks. Hy3 reports 78.0 on SWE-Bench Verified and lower hallucination rates, and…