Chinese AI giant Zhipu AI has unveiled GLM-5.2: 1 million tokens of context and 753 billion parameters — a new standard for open-source models.
Chinese startup Zhipu AI has officially unveiled its new flagship language model — GLM-5.2. This is not just another update, but a significant step forward in the segment of long agentic tasks and programming. The open-source model features a context window of 1 million tokens and is distributed under the MIT license, opening up broad possibilities for local deployment.
According to the documentation on Hugging Face, GLM-5.2 is designed for text generation in English and Chinese. Its architecture includes 753 billion parameters — making it one of the largest open-source models on the market.
Architectural Innovations and Performance
A key feature of GLM-5.2 is support for multiple levels of "reasoning intensity." This allows users to flexibly balance between response quality and latency. The architecture integrates IndexShare components and an updated MTP layer for speculative decoding. Developers claim that IndexShare reuses one indexer for every four layers of sparse attention, reducing operations per token by 2.9 times. Meanwhile, the MTP update increases confirmation length by up to 20%.
Test results are impressive. In three key benchmarks — FrontierSWE, PostTrainBench, and SWE-Marathon — 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.
Availability and Ecosystem
GLM-5.2 is distributed under the MIT license, making it as accessible as possible for developers. For local deployment, support is announced for popular frameworks: SGLang, vLLM, Transformers, KTransformers, and Docker Model Runner. Additionally, quantizations are available for llama.cpp, Ollama, and LM Studio.
It is worth noting that Zhipu AI continues to aggressively expand its presence in the open-source AI market. While major Western players are increasingly closing their models, Chinese developers, on the contrary, are betting on openness and accessibility. GLM-5.2 is not just a technical achievement, but a strategic move that could significantly change the balance of power in the AI ecosystem.
Expert opinion: The emergence of GLM-5.2 with a 1 million token context is a signal that the race for long context is just beginning. For programming tasks and analyzing large documents, this is a breakthrough, but the key question remains the real-world effectiveness of such an architecture in practice. If Zhipu AI manages to confirm the stated metrics, we will witness the formation of a new standard for open-source models.