GLM-5.2 from Zhipu AI: a context window of 1 million tokens and a breakthrough in programming benchmarks
Chinese startup Zhipu AI has officially unveiled its new flagship model — GLM-5.2. This is an open-source solution designed for long agent tasks and programming. A key feature is a context window of up to 1 million tokens, enabling the processing of massive data arrays without quality loss.
The model has 753 billion parameters and is designed for text generation in English and Chinese. Developers emphasize that GLM-5.2 is distributed under the permissive MIT license, allowing for local deployment and customization.
The architecture includes several unique mechanisms. In particular, the IndexShare technology reuses one indexer for every four layers of sparse attention, reducing operations per token by 2.9 times. An updated MTP layer for speculative decoding increases confirmation length by up to 20%, directly impacting generation speed.
In performance tests, GLM-5.2 demonstrated impressive results. In three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — the model outperformed all other open-source solutions. In standard programming tests, it also took the leading position among open models.
For local deployment, support is announced for SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Quantizations are available for llama.cpp, Ollama, and LM Studio, making the model flexible for use on various hardware.
Expert opinion: GLM-5.2 is a significant step forward for the open-source AI segment. The 1 million token context and improved attention architecture make it competitive not only in programming tasks but also in analyzing long documents, which is critical for corporate and financial applications. However, the key challenge remains the efficient local deployment of a model of this scale — 753 billion parameters require significant computing resources.