tencent/Hy3
Tencent released Hy3, a 295B-parameter Mixture-of-Experts model with 21B active parameters, licensed under Apache 2.0. The model outperforms similar-size models and rivals flagship open-source models with 2-5x parameters, supporting a 256K context length. It is available for free on OpenRouter until July 21st.
Background
- **Tencent** is a Chinese multinational conglomerate (owns WeChat, QQ, and major gaming/Tech investments). Hy3 is their latest open-weight large language model.
- **295B total parameters, 21B active**: Hy3 uses a Mixture-of-Experts (MoE) architecture — only a fraction of its 295 billion parameters are "active" per token, making it far more efficient than a dense model of similar total size.
- **Apache 2.0 license**: A permissive open-source license; anyone can download, modify, and use the model commercially without paying Tencent.
- **Context length of 256K tokens**: The model can process roughly 190,000 words of text at once (e.g., an entire long novel or large codebase) in a single pass.
- **Why it matters**: Chinese AI labs (Tencent, Alibaba/Qwen, DeepSeek) have been releasing competitive open-weight models that rival or beat leading US models (Llama, GPT). Hy3 claims to outperform models with 2–5x its active parameter count, meaning it's unusually efficient. The permissive license and free OpenRouter access until July 21 let developers test it without heavy hardware.