Tencent has open-sourced Hy3, a 295B-parameter Mixture-of-Experts model developed by the Tencent Hy Team. Following feedback from over 50 product teams on its preview version, the team significantly improved post-training data quality and scaled up reinforcement learning to enhance reasoning, agentic workflows, and long-context capabilities.
- Hy3 outperformed GLM-5.1 in blind tests with 270 experts, scoring 2.67/4 compared to 2.51/4.
- Hallucination rates dropped from 12.5% to 5.4%, and commonsense error rates fell from 25.4% to 12.7% through fine-grained data cleaning.
- Multi-turn intent tracking issues decreased from 17.4% to 7.9%, with long-dialogue benchmark scores rising from 42.9% to 75.1%.
- Tool-call stability improved, with accuracy variance across different agent scaffoldings remaining within 4% on SWE-Bench Verified.
These improvements position Hy3 as a reliable and cost-effective option for productivity tasks such as coding, document processing, and financial analysis.