GLM-5.2 from Zhipu AI: 1 million tokens of context and a new era of open language models
Chinese startup Zhipu AI has unveiled its latest flagship language model, GLM-5.2, designed to tackle long-horizon agent tasks and complex programming challenges. This open-source solution, distributed under the MIT license, offers a context window of 1 million tokens — a significant step forward for local deployment.
The model, available on Hugging Face, boasts 753 billion parameters and supports text generation in English and Chinese. A key feature of GLM-5.2 is the ability to adjust "reasoning intensity," allowing developers to choose between response quality and latency.
The model architecture includes innovative components: IndexShare and an updated MTP layer for speculative decoding. IndexShare reuses a single indexer for every four layers of sparse attention, reducing operations per token by a factor of 2.9. The MTP update, in turn, increases confirmation length by 20%.
In key benchmarks such as FrontierSWE, PostTrainBench, and SWE-Marathon, GLM-5.2 outperformed all other open-source models. In standard programming performance tests, it also took the leading position among open-source solutions.
For local deployment, the model supports frameworks such as SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Quantizations are available for llama.cpp, Ollama, and LM Studio, making GLM-5.2 a flexible tool for developers.
Expert Commentary: GLM-5.2 is not just another update, but a serious bid for leadership in the open model segment. 1 million tokens of context combined with local deployment capability and the MIT license makes this model extremely attractive for the enterprise sector, where data control and performance are critical. However, given recent scandals regarding the "openness" of models in China, the community should carefully examine how truly independent this solution is.