GLM-4.5

GLM-4.5, released by Zhipu AI (Z.ai) in July 2025, is an open-weights model family built for agentic, reasoning, and coding (ARC) tasks. The flagship model has 355 billion total parameters with 32 billion active at inference time using a Mixture-of-Experts architecture, while the lighter GLM-4.5-Air variant has 106 billion total parameters with 12 billion active. Both support 128K token input and up to 96K token output.

GLM-4.5 ranked 3rd globally on aggregate benchmarks, behind only OpenAI's o3 and xAI's Grok 4. In tool-use benchmarks, it achieved 90.6% accuracy, outperforming Claude Sonnet 4 (89.5%), Kimi K2 (86.2%), and Qwen3-Coder (77.1%). On MATH-500, it scored 98.2% (equaling Claude 4 Opus), and on AIME24 it scored 91.0%, beating Claude Opus 4 (75.7%). On SWE-bench Verified, it scored 64.2%.

The models were pre-trained on 22 trillion tokens - 15 trillion of text followed by 7 trillion of code and reasoning data. The weights are released under the MIT license, making them freely available for commercial use. GLM-4.5 is accessible via API at $0.60/$2.20 per million input/output tokens, with the Air variant at $0.20/$1.10.