Tencent Releases Hy3: 295‑Billion‑Parameter MoE Model Challenging Trillion‑Scale Leaders

Tencent has made its Hy3 model publicly available on Hugging Face. Hy3 features a 295‑billion‑parameter mixture‑of‑experts (MoE) architecture. The model is positioned as a competitor to trillion‑parameter state‑of‑the‑art systems. Its design allows scaling efficiency by activating only relevant expert subsets. Early benchmarks suggest performance comparable to larger models on several tasks. The release provides researchers access to a high‑capacity model without full compute costs. Tencent highlights the model’s versatility across language and reasoning applications. Community feedback will determine its adoption and further optimization.