tencent/Hy3

Tencent has released Hy3, a large-scale artificial intelligence model developed by the company's Hy Team. It is a Mixture-of-Experts model featuring 295 billion total parameters, with 21 billion active parameters and an additional 3.8 billion parameters dedicated to its MTP layer. The model supports a context length of 256,000 tokens, allowing it to process extensive amounts of information at once.

The development of Hy3 followed a preview phase that incorporated feedback from over 50 internal products and utilized improved training data. According to the release details, the model performs better than other models of a similar size and competes with flagship open-source models that are two to five times larger. It has demonstrated increased utility across a variety of productivity tasks and software applications.

The full-sized version of the model is available on Hugging Face at 598 gigabytes, with a compressed FP8 quantized version offered at 300 gigabytes. The model is released under an Apache 2.0 license, making it accessible for broader use. Users can access the model for free on the OpenRouter platform until July 21st. This release represents a significant expansion of Tencent's open-source AI offerings, providing high-performance capabilities for developers and researchers.