Simon Willison
tencent/Hy3
context
What happened
Tencent has released Hy3, a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters. It features a 256K context length, is licensed under Apache 2.0, and is currently available for free testing on OpenRouter.
Why it matters
The release of Hy3 expands the options for developers seeking large-scale, open-weights MoE models with long-context capabilities.
The take
Another massive open-weights MoE model enters the ecosystem. While the Apache 2.0 license and 256K context are great, it represents incremental progress rather than a paradigm shift in how we build LLM systems.
Do this
Test Hy3 on OpenRouter if you are looking for an Apache 2.0 licensed alternative to existing open-weights MoE models for long-context tasks.
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.