Tencent opens full Hy4 flagship weights, FP8 checkpoint and pretrained base
The 770B-backbone multimodal MoE is now downloadable, but its 49B active-parameter figure, separate vision and…
Read storyDAILY EDITION
3 source-linked reports.
Follow Open Model Weights on Google See more of our open-weight AI reporting in Google Search.
The 770B-backbone multimodal MoE is now downloadable, but its 49B active-parameter figure, separate vision and…
Read storyFour Chinese-English models from 2B dense to 35B-A3B MoE process text, images, audio and video together…
Read storyThinking Machines Lab offers Apache-2.0 weights in a roughly 20GB mixed-precision checkpoint, reusing middle…
Read story