GLM-5.2 from Zhipu AI: 1 million tokens and 753 billion parameters — a new standard for open AI models
Chinese startup Zhipu AI has released its new flagship language model, GLM-5.2. This open-source solution, focused on long-term agent tasks and programming, is now available under the MIT license. Its key feature is a context window of 1 million tokens, enabling the processing of large data arrays without loss of coherence.
The model, uploaded to Hugging Face, is designed for text generation in English and Chinese. Its size is an impressive 753 billion parameters. GLM-5.2 offers several levels of "reasoning intensity," giving users the flexibility to balance between response quality and latency time.
The architecture incorporates two key innovations: 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 2.9 times. The MTP update, in turn, increases the confirmation length by up to 20%.
The test results speak for themselves. 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.
The model is distributed under the open MIT license, removing many legal barriers for developers. For local deployment, support is claimed for SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Additionally, quantizations are available for llama.cpp, Ollama, and LM Studio, making GLM-5.2 accessible to a wide range of hardware configurations.
My analysis: GLM-5.2 is a significant step forward for open-source AI. The 1 million token context and 753 billion parameters put it on par with the best proprietary models, and the MIT license makes it ideal for integration into commercial products. In the coming months, we will likely see a wave of solutions based on this model, especially in the areas of code automation and large text analysis.