GLM-5.2 from Zhipu AI: 1 million token context and open-source code for a breakthrough in programming
Chinese startup Zhipu AI has introduced its new flagship language model GLM-5.2, designed for long-term agent tasks and programming. This is a fully open-source solution with a context window of 1 million tokens, distributed under the MIT license and supporting local deployment.
The model, available on Hugging Face, has 753 billion parameters and is intended for text generation in English and Chinese. A key feature of GLM-5.2 is the flexible adjustment of "reasoning intensity," allowing users to choose a balance between response quality and latency.
The model architecture includes two important innovations: the IndexShare mechanism and an updated MTP layer for speculative decoding. IndexShare allows reusing one indexer for every four layers of sparse attention, reducing operations per token by 2.9 times. The MTP update, in turn, increases the confirmation length to 20%.
Test results are impressive. In three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — GLM-5.2 outperformed all other open-source models. In standard programming performance tests, it also took a leading position among open-source solutions.
GLM-5.2 is distributed under the open MIT license. For local deployment, support is claimed for frameworks such as SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Quantizations for llama.cpp, Ollama, and LM Studio are also available.
Expert opinion: The release of GLM-5.2 is a significant step forward for open-source AI. The 1 million token context opens up new possibilities for analyzing large volumes of code and documentation, and the MIT license makes the model accessible to any developer. It is especially important that Zhipu AI is not just increasing "raw" performance but also implementing mechanisms that reduce computational costs — this is a key trend for democratizing access to powerful AI models.