🤖 Tencent has introduced Hy3 — a powerful MoE model with 295B parameters (21B active).
The model is designed for reasoning, coding, and agentic tasks, supporting a context window of up to 256K tokens. Thanks to post-training optimization, the hallucination rate has been reduced from 12.5% to 5.4%.
🌍 The release of a highly efficient MoE model of this scale expands the possibilities for creating complex autonomous agents that can work with massive volumes of data at relatively low computational costs.
👤 Users can now utilize a more powerful and reliable AI assistant for programming and data analysis that makes fewer factual errors and understands instructions better.
Source 1: https://github.com/Tencent-Hunyuan/Hy3 Source 2: https://huggingface.co/tencent/Hy3