Chinese startup Zhipu AI has released its flagship language model GLM-5.2 for long agentic tasks and programming. The open-source solution has a context window of 1 million tokens, an MIT license and support for local deployment.
On its Hugging Face card, the model is listed as a text-generation model for English and Chinese. Its size is 753 billion parameters.
GLM-5.2 supports several levels of "reasoning intensity" to let users choose between quality and latency. The architecture also incorporates IndexShare and an updated MTP layer for speculative decoding.
According to the developers, IndexShare reuses a single indexer for every four layers of sparse attention and reduces the number of operations per token by a factor of 2.9. The MTP update increases the verification length by up to 20%.
In three key benchmarks — FrontierSWE, PostTrainBench and SWE-Marathon — GLM-5.2 outperformed other open-source models.
In standard programming performance tests, GLM-5.2 also became the most powerful open-source model.
GLM-5.2 is distributed under the open MIT license. For local deployment, support is announced for SGLang, vLLM, Transformers, KTransformers and Docker Model Runner. Quantizations are available for llama.cpp, Ollama and LM Studio.
As a reminder, in June the Rio de Janeiro IT company IplanRIO presented Rio 3.5 Open 397B as an open AI model trained with public funds. However, a day later the Nex team stated that the tool looks like a direct merge of Nex-N2-Pro and Qwen3.5-397B-A17B.
Source: ForkLog
