tencent/Hy3

This article covers the release of Tencent's Hy3, a 295B-parameter Mixture-of-Experts (MoE) language model with 21B active parameters and 3.8B MTP layer parameters. Developed by the Tencent Hy Team, Hy3 builds on feedback from 50+ products and improved post-training data. It outperforms similar-size models and rivals flagship open-source models with 2-5x more parameters, showing gains in productivity tasks. The model has a 256K context length, is available as a 598GB full-size or 300GB FP8 quantized version on Hugging Face, and is free on OpenRouter until July 21st. The article also includes a test generating an SVG of a pelican riding a bicycle.