Tencent has released and open-sourced Hy4 preview, the next-generation large language model from its Hunyuan team. The mixture-of-experts model packs 770 billion total parameters with 49 billion active per token, and supports a context window of more than one million tokens. Weights are live on Hugging Face, ModelScope, GitCode and CNB under the Apache 2.0 license, including an FP8 quantized build for cheaper deployment.
Hy4 preview targets real-world productivity rather than pure benchmark chasing. Tencent says it excels across coding, office work, data analysis, game development and scientific research. In an internal blind evaluation with 163 experts across 203 engineering tasks it scored 2.99 out of 4.00, edging GLM-5.3 at 2.92 and Kimi K3 at 2.94. The model also helped optimize its own training and inference pipelines, lifting end-to-end throughput by 31.8 percent in what Tencent calls an early recursive self-improvement loop.
Access is broad from day one. Hy4 preview is available in WorkBuddy, CodeBuddy, Yuanbao and ima, with two weeks of free use on WorkBuddy and CodeBuddy. API access runs through Tencent Cloud TokenHub and OpenRouter at 0.834 dollars per million input tokens and 2.501 per million output tokens. For founders, a frontier-class open model with a 1M-token context at that price is another sign that open-weight models are closing the gap with their closed rivals.