Cambricon(SH:688256) Zheng Quan Shi Bao·2025-09-30 09:16Core Insights - The release of the new generation large model GLM-4.6 by the domestic unicorn company Zhipu marks a significant advancement in programming capabilities, surpassing the latest model DeepSeek-V3.2-Exp and aligning with global leader Claude Sonnet 4 in various assessments [1][2][5] Model Performance - GLM-4.6 has shown substantial improvements in core capabilities such as Agentic Coding, long context processing, reasoning ability, information retrieval, and text generation, making it the strongest coding model in China [2][5] - The context window has been increased from 128K to 200K, allowing for better handling of longer code and agent tasks [5] - The model has improved reasoning capabilities and supports tool invocation during reasoning processes, enhancing its overall performance [5] Token Efficiency - GLM-4.6 has achieved over 30% savings in token consumption compared to its predecessor GLM-4.5, indicating improved efficiency in processing tasks [5] Open Source and Global Positioning - GLM-4.6 is set to be open-sourced on platforms like Hugging Face and ModelScope under a permissive MIT license, positioning it as one of the strongest general-purpose models in the global open-source ecosystem [6] Chip Adaptation and Ecosystem Development - The model has been adapted for use on domestic AI chips from companies like Cambricon and Moore Threads, marking a significant step in the collaboration between domestic large models and chips [7][8] - The successful adaptation of GLM-4.6 on these chips demonstrates the capability of domestic GPUs to collaborate with cutting-edge large models, accelerating the development of a self-controlled AI technology ecosystem [7][8] Industry Collaboration - The rapid adaptation of GLM-4.6 following the release of DeepSeek-V3.2-Exp indicates deep collaboration among leading companies in the Chinese AI industry, suggesting a shift towards a unified ecosystem of software and hardware [8] - Zhipu has initiated A-share listing guidance, aiming to become the first publicly listed company focused on domestic AI large models, signaling a transition from technological competition to commercialization and capital operation [8]