AI Intelligence // signal over noise
← back to feed
Medium LLM

Tencent Hy3 Review: 21B-Active Open MoE That Beat GLM-5.1 in a Blind Test

reasoning
What happened
Tencent released Hy3, an open MoE model with 295B total and 21B active parameters, licensed under Apache 2.0 with a 256K context window. In a blind evaluation conducted by 270 domain experts against Zhipu's GLM-5.1, Hy3 won with a score of 2.67 to 2.51 on a 4-point scale. The model is immediately available on OpenRouter and Tencent Cloud.
Why it matters
It expands the pool of highly capable, open-weights MoE models suitable for cost-sensitive, long-context applications.
The take

Tencent's shift toward open-weights frontier competition is notable. While not a massive leap in capabilities, a highly competitive 21B active parameter MoE with a 256K context window under Apache 2.0 provides an excellent, cost-effective alternative for developers targeting bilingual (Chinese/English) agent workflows.

Do this
Evaluate Tencent Hy3 on OpenRouter as a potential open-weights alternative for long-context reasoning tasks.
Read the source →

Don't read this site daily. Get it in your inbox.

The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.