Tencent Launches Hunyuan 3: 295B MoE Beats GLM-5.1 in Blind Test

kimmonismus · x · 2026-07-06

Reportedly, Tencent has released Hunyuan 3 (Hy3), utilizing a 295B parameter MoE architecture, 21B activated parameters, and a 256K context length, beating GLM-5.1 in blind tests. Notably, it achieves this efficiency using standard GQA rather than sparse attention or MLA, indicating further room for architectural optimization. This news was published by a third party and remains officially unconfirmed.

Related event: Tencent Open-Sources Hunyuan Hy3 for Agentic Workloads(26 posts)→

Original post →

More from Models

Models channel →