Tencent previews Hy4, a 770B-parameter model, against GLM-5.3 and Kimi K3

A report from shattered.io says Tencent has previewed Hy4, a large model with 770 billion parameters. The headline places Hy4 directly alongside GLM-5.3 and Kimi K3, framing it as part of a new round of head-to-head comparisons among large models.

The available information is limited: the model is named Hy4, it is at a preview stage, its parameter count is given as 770B, and its comparison targets are GLM-5.3 and Kimi K3. Beyond that, the candidate material does not disclose training data, context length, openness or specific capabilities.

Against the current industry rhythm, parameter counts in the 700B range have become common among leading models, and preview releases are typically used to gather feedback ahead of a full launch.

Placing GLM-5.3 and Kimi K3 as the comparison points also suggests that evaluation now depends on head-to-head results against contemporaneous rivals rather than a single vendor’s claims.

It is worth noting that the verifiable information currently stops at the headline level; no complete benchmark data or official technical documentation has been seen, so judgments about Hy4’s real capabilities should stay cautious.

What to watch next: whether Tencent publishes fuller model cards and benchmark results, how large the gap between the preview and the final release turns out to be, and how these models compare on the same set of tasks.