Hy3 – 295B MoE model (21B active, 192 experts) (twitter.com)

Tencent has launched Hy3, an impressive model boasting 295 billion parameters with a mixture of experts (MoE) architecture featuring 21 billion active parameters and 192 experts. This model is positioned as a competitive solution against existing trillion-scale language models, promising enhanced reliability and affordability for a variety of agentic use cases. Hy3’s open-source approach under the Apache 2.0 license makes it appealing for commercial use, further broadening accessibility in the AI/ML ecosystem. The introduction of Hy3 is significant as it demonstrates the potential for advanced AI models to become more resource-efficient while maintaining high performance, which is crucial for developers and researchers in the field. Offering a free API for the next two weeks allows users to experiment with its capabilities without financial commitment, fostering innovation and engagement within the community. By integrating powerful tools and flexibility, Hy3 could help accelerate the development of intelligent applications and broaden the horizons for future AI research.