GLM 4.6 by Zhipu is Here: Domestic Chips Join Forces to Drive AI Advancement

In recent tech updates, Zhipu Company announced the latest version of its GLM series - GLM-4.6. The release of this new version marks another breakthrough in domestic chip technology. According to official information, GLM-4.6 uses advanced domestic chips from Cambrian, successfully achieving a mixed quantization deployment of FP8+Int4, which is a first for domestic chips. Notably, this innovation maintains model accuracy while significantly reducing inference costs, opening up a new path for local operation of large models on domestic chips.

The release of GLM-4.6 signifies further enhancement of China's independent R&D capabilities in the AI field. Combined with Cambrian and MoLeiXianChen, it demonstrates the strong performance of GLM-4.6 under the vLLM inference framework. The new generation GPU from MoLeiXianChen also runs stably with native FP8 precision. All of this verifies the ecological compatibility and rapid response capability of the MUSA architecture.

Zhipu will officially launch the GLM-4.6 service through its MaaS platform to both individual and enterprise users. This new version is not only a technological upgrade but also includes enhanced image recognition and search capabilities, supporting various mainstream programming tools such as Claude Code, Roo Code, and Kilo Code. Meanwhile, Zhipu's GLM Coding Max package provides more usage space for frequent developers, with a monthly fee as low as 20 yuan.

As technology continues to evolve, existing subscription users of GLM Coding Plan will automatically upgrade to GLM-4.6. The introduction of enterprise packages is specifically tailored for enterprises that require security, cost-effectiveness, and top-tier international performance. In the future, domestically developed GLM large models will work together with domestic chips to optimize performance and efficiency in model training and inference processes.

The release of GLM-4.6 not only showcases Zhipu's technical strength in the AI field but also injects new momentum into the development of domestic chips. With the maturation of technology, domestic large models will demonstrate their capabilities in more application scenarios.