Models

Tencent Hy4

1 synchronized story carry this tag.

Tencent publishes flagship Hy4 instruct, FP8 and base weights

On 20 November Tencent made the flagship Hy4 instruct model, a quantized FP8 instruct variant and the pretrained base checkpoint public, after the 4 November hosted launch. The earlier Hy4 preview and the smaller Hy4 Flash are separate artifacts. The flagship release uses the Tencent Hy License.

Why it matters: Teams can now evaluate a specific full flagship checkpoint locally rather than using only a hosted endpoint or substituting the preview weights. Separate base and instruction-tuned variants support different kinds of evaluation.

Read storyOriginal source · tencent.com