Chinese AI giant Zhipu AI releases GLM-5.2: 1 million tokens of context and 753 billion parameters
Chinese startup Zhipu AI has officially unveiled its new flagship language model, GLM-5.2, which is now available as open source. This solution is designed for long agentic scenarios and programming, capable of handling a context window of up to 1 million tokens. The model is distributed under the MIT license, opening up broad opportunities for local deployment.
According to data on Hugging Face, GLM-5.2 is a generative model optimized for English and Chinese languages. Its total parameter count stands at an impressive 753 billion, making it one of the largest open models on the market.
A key feature of GLM-5.2 is its support for multiple levels of "reasoning intensity," allowing users to flexibly balance response quality and latency. The architecture integrates innovative components: the IndexShare mechanism, which reuses a single indexer for every four layers of sparse attention, and an updated MTP layer for speculative decoding. Developers claim that IndexShare reduces operations per token by 2.9 times, while the MTP update increases confirmation length by up to 20%.
In key benchmarks such as FrontierSWE, PostTrainBench, and SWE-Marathon, GLM-5.2 confidently outperformed all other open-source models. In standard programming performance tests, it also took the leading position among open-source solutions.
For local deployment, GLM-5.2 supports frameworks such as SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Additionally, quantizations are available for llama.cpp, Ollama, and LM Studio, making the model accessible to a wide range of developers.
Expert commentary: The release of GLM-5.2 is not just another launch but a serious signal that China is accelerating its pace in the open AI model race. A context of 1 million tokens and 753 billion parameters represent a level that seemed unattainable for open source until recently. The focus on agentic tasks and programming is particularly impressive: it is a direct challenge to the positions of Western giants, which have yet to offer similar open-source solutions. The market is clearly entering a new phase of competition, where the key factor is not just model size, but its practical applicability and deployment speed.