Zhipu's GLM-5.2 Tops Open-Source Models in AA Benchmark With 51 Points, Matches GPT-5.5 on GDPval Test

According to Beating AI monitoring, Zhipu's latest GLM-5.2 model ranks first among open-source models in Artificial Analysis' intelligence benchmark v4.1 with a score of 51 points, surpassing MiniMax-M3 (44 points) and DeepSeek V4 Pro (44 points). In the GDPval-AA v2 test simulating real-world knowledge work, GLM-5.2 achieved 1,524 points (human baseline: 1,000), matching closed-source frontier model GPT-5.5 on xhigh reasoning mode and outperforming MiniMax-M3 (1,418 points) and DeepSeek V4 Pro (1,328 points).

The model features 744B total parameters with 40B active parameters and expands context window from 200K to 1M tokens. GLM-5.2 operates under MIT license and is deployed across multiple platforms including Zhipu's official API, SiliconFlow, and DeepInfra, with average task cost at approximately $0.46 per task.