Intel / post

Hy3 lands: 295B MoE under Apache 2.0, free API for two weeks

Source: x.com
Date: Published Jul 6, 2026

Context

Tencent's official Hy3 release followed the Hy3 preview with further training informed by product feedback. The model card describes a model with 295 billion parameters, of which 21 billion are active for each text . This design uses a selected portion of the model's capacity for each processing step.

Hy3's downloadable weights are released under Apache 2.0, providing a route for developers to run and adapt the model. The launch announcement also offered a two-week API trial; that was a launch promotion, rather than an ongoing promise of free access.

Terms in this piece · Glossary

mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Read the source x.com