Chinese AI giant Zhipu AI has unveiled GLM-5.2: a 1 million token context and open-source code.
Chinese startup Zhipu AI has officially released its latest flagship language model, GLM-5.2, aimed at solving complex agent tasks and programming. Its main feature is a context window of up to 1 million tokens, allowing the model to process vast amounts of data in a single pass. The model is distributed as open source under the MIT license and supports local deployment.
According to information on Hugging Face, GLM-5.2 boasts a colossal 753 billion parameters and is optimized for text generation in English and Chinese. This makes it one of the largest open models on the market.
Architectural Innovations and Performance
GLM-5.2 implements a unique "reasoning intensity" system, allowing users to dynamically choose a balance between response quality and latency. The architecture also integrates IndexShare mechanisms and an updated MTP layer for speculative decoding. Developers claim that IndexShare reuses a single indexer for every four layers of sparse attention, reducing operations per token by 2.9 times, while the MTP update increases confirmation length by up to 20%.
Test results are impressive: in three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — GLM-5.2 surpassed all other open-source models. In standard programming performance tests, it also took a leading position among open-source solutions.
Openness and Accessibility
GLM-5.2 is distributed under the MIT license, providing maximum flexibility for developers. For local deployment, support for popular frameworks is announced: SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Additionally, quantizations for llama.cpp, Ollama, and LM Studio are available, simplifying integration into various environments.
My expert opinion: The release of GLM-5.2 is a significant step by Zhipu AI in the race for leadership in the open AI model segment. The 1 million token context and support for local deployment make this model extremely attractive for the enterprise sector, especially in the areas of programming automation and long document processing. However, it is worth noting that against the backdrop of recent scandals with "open" models, where some turned out to be direct merges of other projects, the community should carefully examine the originality of the GLM-5.2 architecture.