๐Ÿค– GLM-5.2 has arrived, and it's open-weights from the start

The flagship GLM-5.2 model has been introduced with open weights under the MIT license. Key features: a stable 1 million token context window and the IndexShare architecture, which reduces computational load by 2.9x when working with long texts.

๐ŸŒ The emergence of a powerful open-source model with such a large context lowers the barrier to entry for creating autonomous AI agents capable of working with entire code repositories.

๐Ÿ‘ค The model can now be run locally (via vLLM, SGLang, etc.) or through an API, which is critical for deep code analysis and complex systems.

Source 1: http://z.ai/blog/glm-5.2 Source 2: http://huggingface.co/zai-org/GLM-5.2