tencent/Hy3
Simon Willison · Simon Willison · 2026-07-06
Tencent releases Hy3, a 295B-parameter Apache 2.0-licensed mixture-of-experts language model with 21B active parameters and 256K context length, available free on OpenRouter through July 21, 2026.
Appears in
Extraction
Topics: llm-releasemixture-of-expertsopen-source-aiai-in-chinagenerative-ai
Claims
- Hy3 is a 295B-parameter MoE model with 21B active parameters developed by the Tencent Hy Team, following a preview launch in late April 2026.
- Hy3 outperforms similar-size models and rivals flagship open-source models with 2-5x more parameters.
- The full-sized model weighs 598GB on Hugging Face, with an FP8 quantized version at 300GB and a 256K token context length.
- The model was refined using feedback from 50+ internal Tencent products and scaled with higher-quality post-training data.
- Hy3 is available for free on OpenRouter until July 21, 2026.
Key quotes
Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team.
Hy3, which outperforms similar-size models and rivals flagship open-source models with 2-5x parameters.
It's available for free on OpenRouter until July 21st.