Tencent releases Hy3, an open-source model competing with AI 5x its size
Tencent introduces Hy3, an open-source language model with 295B parameters (only 21B active at once) using a mixture-of-experts architecture.
Published 1sem1 sourceNotable
Lire en français
≈ 23s
The fact
Tencent claims Hy3 matches models 2-5x its size while halving hallucination rates to 5.4%.
The model improves computational efficiency without sacrificing performance compared to larger architectures.
Click the link to read an article on the topic: