Intel / post

Tencent opens Hy4 preview model on OpenRouter and TokenHub

Source: x.com
Date: Published Aug 28, 2026

Context

This post is the distribution side of Tencent's Hy4 preview release: it tells people where they can use the model without hosting it. Tencent's August 28 announcement describes Hy4 preview as a 770-billion-parameter model that activates 49 billion parameters per and accepts more than 1 million tokens of , a size that puts running it yourself beyond the hardware most individuals have.

There are two kinds of access. People can try it inside Tencent apps, including the WorkBuddy workspace, the CodeBuddy coding tool and Yuanbao, and Tencent made it free on WorkBuddy and CodeBuddy for two weeks from launch. Developers can call it through an API on Tencent Cloud's TokenHub gateway or on OpenRouter, a service that routes requests to many providers' models. Tencent lists the API price at US$0.834 per million input tokens, US$2.501 per million output tokens and US$0.042 per million tokens served from cache.

The post's claim of "strong performance" rests on Tencent's own testing. Its main number comes from an internal blind in which 163 experts graded 203 engineering tasks: Hy4 preview averaged 2.99 out of 4, against 2.94 for Kimi K3 and 2.92 for GLM-5.3. A margin that small, from a test the vendor designed and ran, shows the model is in the same range as those competitors rather than clearly ahead of them. The free two-week window is the more useful fact for anyone deciding whether to test it on their own work.

Terms in this piece · Glossary

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.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

Read the source x.com