Chinese AI giant Zhipu has unveiled GLM-5.2: 753 billion parameters and a context of 1 million tokens
The open-source language model market has gained a powerful player. Chinese startup Zhipu AI has officially unveiled its flagship development — the GLM-5.2 model, designed for long agentic scenarios and programming tasks. The key feature of the new model is a context window of 1 million tokens, opening up possibilities for processing vast amounts of data without losing coherence.
The model, available on Hugging Face, boasts a colossal 753 billion parameters and supports text generation in English and Chinese. GLM-5.2 is distributed under the open MIT license, allowing developers to integrate it into their projects without restrictions, including local deployment.
The GLM-5.2 architecture includes several innovative solutions. First, it implements support for multiple levels of "reasoning intensity," giving users flexibility in choosing between response quality and latency. Second, it incorporates the IndexShare mechanism, which reuses one indexer for every four layers of sparse attention, reducing operations per token by 2.9 times. Additionally, the updated MTP layer for speculative decoding increases the confirmation length to 20%.
Test results are impressive. In three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — GLM-5.2 outperformed all existing open-source models. In standard programming performance tests, it also took a leading position among open-source counterparts.
For local deployment, the model supports SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Quantizations for llama.cpp, Ollama, and LM Studio are also available, making it accessible to a wide range of developers.
Expert opinion: The emergence of GLM-5.2 with a 1 million token context and an open MIT license is a significant step forward for the industry. Chinese developers continue to actively expand their presence in the open-source AI model segment, offering competitive solutions that can rival Western counterparts. It is especially valuable that Zhipu AI focuses on agentic tasks and coding — these are precisely the areas currently driving growth for the enterprise sector.