Morgan Stanley: Tencent (00700) gives an "overweight" rating to Hy3 Preview.

Morgan Stanley released a research report stating that TENCENT (00700) released and open-sourced the next-generation large language model, Hy3 Preview, on the 23rd of yesterday. It is positioned as the first step in rebuilding the Hybrid (HY) basic model. The model is based on a Mixture of Experts (MoE) architecture, integrating fast and slow thinking, with a total parameter count of 295 billion (21 billion activation parameters), a context window of 256K. It has seen improvements in complex reasoning, instruction following, context learning, code generation, agentability, and inference performance. The firm has given Tencent a "hold" rating with a target price of 650 Hong Kong dollars. The firm stated that Hy3 Preview has made significant improvements in code generation and agentability, with a SWE-Bench Verified score of 74.4% (compared to 53% for Hy2) which is roughly on par with GLM 4.7's 73.8%. Tencent's comprehensive optimization provides the best cost-performance ratio, with a 40% increase in inference efficiency and a significant decrease in overall costs. The model has been integrated into Tencent products such as Yuanbao, CodeBuddy, WorkBuddy, QQ, ima, and offers competitive API pricing and customized Token plans on Tencent Cloud, with a monthly fee for the personal version as low as 28 RMB. Morgan Stanley points out that Tencent plans to collect feedback from developers and users through open sourcing to improve the formal version, expand training and reinforce learning, deepen product ecosystem integration, and pursue differentiated model capabilities to enhance real-world applicability.