Chinese AI giant Zhipu AI releases GLM-5.2: 1 million token context and open-source code
Chinese startup Zhipu AI has officially unveiled its new flagship language model, GLM-5.2. This open-source solution, designed for long-term agent tasks and programming, features a context window of 1 million tokens. The model is distributed under the MIT license, opening up broad possibilities for local deployment.
According to specifications on Hugging Face, GLM-5.2 is a text generation model for English and Chinese, with 753 billion parameters. The developers have incorporated multiple levels of "reasoning intensity" into the architecture, allowing users to flexibly choose between response quality and latency. Key innovations include the IndexShare mechanism and an updated MTP layer for speculative decoding.
IndexShare, according to the team, reuses a single indexer for every four layers of sparse attention, reducing operations per token by a factor of 2.9. Meanwhile, the MTP update increases confirmation length by 20%. These optimizations make the model not only more performant but also more computationally economical.
In tests, GLM-5.2 demonstrates impressive results. In three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — the model outperformed all other open-source solutions. Moreover, in standard programming performance tests, it also took the leading position among open models.
For local deployment, GLM-5.2 supports SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Quantizations for llama.cpp, Ollama, and LM Studio are also available. This makes the model accessible to a wide range of developers and researchers.
My expert opinion: The release of GLM-5.2 is a significant signal for the market. Zhipu AI is not just catching up with Western competitors but is offering unique technical solutions, such as IndexShare, which could change the economics of working with large language models. The open MIT license and support for local deployment make this model particularly attractive for the corporate sector, where data security and infrastructure control are paramount. While some projects, like the recent Rio 3.5, face accusations of plagiarism, Zhipu AI demonstrates independent and deep engineering work.