Intel / post

✔️Hy4 preview just shipped an upgrade.

Source: x.com
Date: Published Sep 7, 2026

Context

Hy4 preview is Tencent Hy's large coding-and-, which reached tools like Cline earlier this year running at 770 billion parameters with 49 billion active. In this update, Tencent says it addressed user reports of the model over-thinking and over-verifying on complex tasks, resulting in fewer turns and lower input and output counts for the same task quality, based on both and human .

No specific benchmark scores or token-count figures accompanied the claim, so the reduction should be read as Tencent's own characterization of an internal comparison rather than an independently verifiable measurement.

Terms in this piece · Glossary

reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
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.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
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