Crypto news

Chinese company Zhipu AI has officially released its flagship language model GLM-5.2, designed to solve complex agent tasks and programming. This open-source solution, distributed under the MIT license, is already available for local deployment and boasts an impressive context window of 1 million tokens.

According to data on Hugging Face, GLM-5.2 is a model for text generation in English and Chinese, with a colossal 753 billion parameters. The architecture is based on innovative IndexShare mechanisms and an updated MTP (Multi-Token Prediction) layer designed for speculative decoding.

Zhipu AI engineers claim that IndexShare allows reusing one indexer for every four layers of sparse attention, reducing the number of computational operations per token by 2.9 times. In turn, the MTP update increases confirmation length by 20%, significantly accelerating the generation process.

Breakthrough in Benchmarks

In key performance tests such as FrontierSWE, PostTrainBench, and SWE-Marathon, GLM-5.2 outperformed all other open-source models. In standard programming tests, it also took the leading position among open-source counterparts.

The model supports multiple levels of "reasoning intensity," allowing users to flexibly choose between response quality and latency. For local deployment, compatibility with SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner is claimed. Additionally, quantizations for llama.cpp, Ollama, and LM Studio are available.

Expert opinion: The release of GLM-5.2 is not just another launch but a signal that the race in the open-source AI segment is moving to a new level. The 1 million token context and focus on agent tasks make this model a serious tool for developers, especially in the field of automating complex business processes. However, it is worth remembering that such large-scale models require significant computational resources, and the real advantage over competitors will depend on the efficiency of local deployment.