Cambricon(SH:688256) 证券时报·2025-09-30 09:09Core Viewpoint - The release of the new generation model GLM-4.6 by Zhiyu marks a significant advancement in programming capabilities, positioning it as the leading domestic coding model and enhancing the collaboration between domestic AI models and chips [1][2][14]. Group 1: Model Performance - GLM-4.6 has achieved substantial improvements in various capabilities, including real programming, long context processing, reasoning ability, information retrieval, and writing skills, surpassing the performance of the latest DeepSeek model [1][12]. - In tests conducted in the Claude Code environment, GLM-4.6 outperformed Claude Sonnet 4 in 74 real-world programming tasks, establishing itself as the strongest coding model in China [12][14]. - The model has improved token efficiency, saving over 30% in token consumption compared to GLM-4.5, while enhancing the usability and aesthetics of front-end code [12][14]. Group 2: Chip Adaptation - GLM-4.6 has been adapted for use with domestic AI chips from Cambrian and Moore Threads, marking a significant step in the integration of domestic models and chips [13][14]. - The model utilizes a mixed quantization inference deployment (FP8+Int4) on Cambrian chips, significantly reducing inference costs while maintaining model accuracy [14]. - The adaptation of GLM-4.6 to Moore Threads' GPU demonstrates the compatibility and rapid adaptation capabilities of domestic chips with advanced models [14][15]. Group 3: Industry Implications - The simultaneous release and adaptation of models by leading domestic companies indicate a deep collaboration within the Chinese AI industry, accelerating the development of a unified ecosystem [15]. - Zhiyu has initiated A-share listing guidance, aiming to become the first publicly listed company focused on domestic AI models, signaling a shift towards commercialization and capital operation in the AI sector [15].