腾讯，重磅开源！

Core Viewpoint - Tencent's Hunyuan-A13B model, an open-source MoE model with 80 billion total parameters and 13 billion active parameters, offers faster inference speed and better cost-effectiveness compared to leading open-source models [1][3][4]. Group 1: Model Performance and Features - Hunyuan-A13B is the first 13 billion parameter MoE open-source model, widely used internally with over 400 business applications and an average daily request exceeding 130 million [3]. - In various authoritative industry benchmarks, Hunyuan-A13B demonstrated competitive performance against models like OpenAI's o1-1217 and DeepSeek's R1-0120, achieving notable scores in mathematics, science, coding, reasoning, and instruction tasks [4][5]. - The model excels in agent tool invocation and long-text capabilities, utilizing a multi-agent data synthesis framework and reinforcement learning for autonomous exploration and learning [5]. Group 2: Strategic Developments and Future Plans - Tencent's recent restructuring of its large model research and development system includes the establishment of new departments focused on large language models and multimodal models, enhancing its technical capabilities [7]. - The company plans to release more models with varying sizes and features to meet diverse enterprise needs, including multimodal foundational models for images, videos, and 3D [6][7]. - Tencent's R&D investment reached 70.69 billion yuan in 2024, with capital expenditures increasing significantly to support AI capabilities and infrastructure, including GPU purchases for inference needs [8].