Tencent has officially launched Hunyuan 3 (Hy3), an AI model featuring a Mixture of Experts (MoE) architecture with 295 billion total parameters. The model boasts a 256K context window and claims to match GPT-5.5 in search capabilities while halving its hallucination rate compared to previous versions. Tencent attributes these improvements to enhanced post-training data quality and expanded reinforcement learning compute scale, with the arrival of professor Yao Shunyu noted as a significant factor. AI
IMPACT This release signifies a significant step for Chinese LLMs, potentially lowering inference costs and challenging top international models in search and RAG applications.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
