GLM-5.2 from Zhipu AI: 1 million token context and architectural innovations for agent tasks
Chinese startup Zhipu AI has unveiled its new flagship language model GLM-5.2, designed for solving long-horizon agent tasks and programming. This open-source solution, distributed under the MIT license, is already available for local deployment and supports a context window of up to 1 million tokens.
The model has 753 billion parameters and is focused on text generation in English and Chinese. One of the key features of GLM-5.2 is support for multiple levels of "reasoning intensity," allowing users to flexibly balance between response quality and latency.
Architectural innovations include the IndexShare mechanism and an updated MTP layer for speculative decoding. The developers claim that IndexShare reuses one indexer for every four layers of sparse attention, reducing operations per token by 2.9 times. In turn, the MTP update increases confirmation length by up to 20%, which is critical for tasks requiring high accuracy and long sequences.
In the FrontierSWE, PostTrainBench, and SWE-Marathon benchmarks, GLM-5.2 confidently outperformed all other open-source models. In standard programming performance tests, it also took a leading position among open-source solutions. This confirms that Zhipu AI has focused on the practical applicability of the model in real-world development scenarios.
GLM-5.2 supports local deployment via SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. For users working with limited resources, quantizations are available for llama.cpp, Ollama, and LM Studio.
My expert opinion: This release is not just another open-source model, but a significant step towards democratizing powerful AI tools. The combination of a 1 million token context with a flexible architecture makes GLM-5.2 a serious competitor to closed solutions, especially in the field of automating complex agent tasks and software development.