New Open-Source AI Model GLM-5.2 Narrows Gap With Top Proprietary Systems

Chinese artificial intelligence company Z AI has released GLM-5.2, a large language model that now holds the top position among all open-weight models on the Artificial Analysis Intelligence Index. The model scored 51 points on the benchmark, placing it ahead of every other downloadable model and closer than ever to leading proprietary systems.

What Makes GLM-5.2 Stand Out

GLM-5.2 is a mixture-of-experts model with 744 billion total parameters, though only 40 billion are active during any single task. This design keeps the model efficient without sacrificing capability. A mixture-of-experts approach means the system activates only the most relevant parts of its network for each request, reducing the computing power needed to generate responses.

The model supports a one million token context window, a major upgrade from the 200,000 tokens available in its predecessor GLM-5.1. For developers and researchers, this means the model can process much longer documents, code repositories, or conversation histories in a single session. Z AI has released GLM-5.2 under the MIT license, allowing anyone to download, modify, and deploy the model freely. It is already available through several inference providers including DeepInfra, Nebius, and Baseten.

Closing the Gap With Proprietary Models

On the Intelligence Index, GLM-5.2's score of 51 places it well above other open-weight competitors like DeepSeek V4 Pro and MiniMax-M3, both of which scored 44. However, a gap remains with proprietary models. Fable 5 leads the overall index at 60 points, while Opus 4.8 and GPT-5.5 sit at 56 and 55 respectively.

GLM-5.2 showed especially strong improvements in scientific reasoning, jumping 16 points on the CritPt benchmark compared to version 5.1. It also gained 16 points on TerminalBench, a test that measures how well models handle real-world coding tasks. These gains suggest Z AI focused on practical performance areas that matter most to developers and researchers.

The release of GLM-5.2 marks another step in the steady march of open-weight models toward proprietary-level performance. At roughly 46 cents per task on the benchmark, it offers strong results at a fraction of the cost charged by closed alternatives. For teams building products powered by large language models, the choice between open and closed systems continues to get harder to make.