Intel / article

Tencent Releases And Open Sources Tencent Hy4 Preview

Source: tencent.com
Author: Tencent
Date: Published Aug 28, 2026

Context

Tencent reports using Hy4 preview to improve the software that runs the model itself. According to its release account, the model analyzed bottlenecks and iterated on optimizations to computation and communication. Tencent reports 31.8% higher end-to-end throughput against its baseline, meaning more model processing completed per unit of time. This is a concrete example of model-assisted infrastructure work. The reported gain concerns serving efficiency; assessing whether it transfers to another deployment requires hardware, workload and baseline details beyond those supplied in the announcement.

Terms in this piece · Glossary

inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
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.

Read the source www.tencent.com