GLM-5.2 from Zhipu AI: 1 million tokens in context and open MIT license
Chinese startup Zhipu AI has unveiled its latest flagship language model, GLM-5.2, designed for long-duration agent tasks and complex programming. This open-source solution, distributed under the MIT license, is already available for local deployment and supports an impressive context window of up to 1 million tokens.
According to the technical documentation on the Hugging Face platform, the model has 753 billion parameters and is designed for text generation in English and Chinese. However, its key feature is a flexible architecture that allows choosing between response quality and latency through several levels of "reasoning intensity."
Architectural Innovations: IndexShare and MTP
GLM-5.2 implements the IndexShare mechanism, which reuses one indexer for every four layers of sparse attention. This reduces the number of operations per token by 2.9 times. Additionally, the updated speculative decoding layer MTP increases the confirmation length by up to 20%, which is critical for tasks requiring high generation speed.
In three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — GLM-5.2 confidently outperformed all other open-source models. In standard programming performance tests, it also took the leading position, confirming its status as the most powerful open-source model to date.
Availability and Deployment
The model is distributed under the open MIT license, allowing its use in commercial and research projects without restrictions. For local deployment, support for popular frameworks is announced: SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Quantizations for llama.cpp, Ollama, and LM Studio are also available, making GLM-5.2 accessible for operation on hardware with limited resources.
Analyst's Opinion: The release of GLM-5.2 with a 1 million token context and an open license is a significant step forward for open-source AI. The ability for local deployment and flexible adjustment of reasoning intensity makes this model particularly attractive for developers of complex agent systems. However, competition in the large language model segment is intensifying, and Zhipu AI will need to confirm its stated performance in practice, especially in real-world usage scenarios.