Zhipu AI has launched GLM-5.2, an open-source language model with a 1 million token context and advanced programming capabilities.
Chinese startup Zhipu AI has launched its latest language model, GLM-5.2, which is tailored for complex tasks and programming applications. This open-source model features a remarkable context window of 1 million tokens and operates under the MIT license, allowing for local deployment.
Available on Hugging Face, GLM-5.2 can generate text in both English and Chinese, boasting an impressive 753 billion parameters.
The model introduces various levels of "reasoning intensity" to optimize the trade-off between output quality and processing speed. Its architecture includes innovations like IndexShare and an upgraded MTP layer to enhance speculative decoding.
Developers note that IndexShare efficiently reuses a single indexer for every four layers of sparse attention, resulting in a 2.9-fold reduction in operations per token. Additionally, the MTP upgrade can extend confirmation length by up to 20%.
In key performance evaluations—FrontierSWE, PostTrainBench, and SWE-Marathon—GLM-5.2 surpassed other open-source models. It also excelled in standard programming performance assessments, making it the leading open-source model in this area.
GLM-5.2 is freely available under the MIT license and supports local deployment through various platforms, including SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. It offers quantization options compatible with llama.cpp, Ollama, and LM Studio.
In related news, IplanRIO, a technology firm from Rio de Janeiro, recently introduced the Rio 3.5 Open 397B as an open AI model funded by public resources. However, shortly thereafter, the Nex team claimed that the model appeared to combine elements from Nex-N2-Pro and Qwen3.5-397B-A17B.
