GLM

GLM is the family of large language models made by Zhipu AI, which presents itself as Z.ai.¹ Zhipu says it built the GLM pre-training framework at the end of 2020 and introduced GLM in March 2021.¹ GLM-130B, an open bilingual English and Chinese model with 130 billion parameters, was open-sourced in September 2022 and described in a paper first posted on October 5, 2022.¹² ChatGLM and the open ChatGLM-6B followed in March 2023, GLM-4 in January 2024, and GLM-4.5, GLM-4.6 and GLM-4.7 in 2025.¹ GLM-5's weights went up on Hugging Face under the MIT License in February 2026.³ The current generation is GLM-5.3, which OpenRouter lists as released on August 18, 2026, a reasoning model for complex software engineering and long-horizon agent tasks whose reasoning is always on, with "low", "high" and "max" effort settings,⁴ and the smaller GLM-5.3-Flash, a native multimodal model released on August 26, 2026; both list a 1,048,576-token context window.⁴⁵

GLM 5 resolves 72.8% of SWE-bench Verified tasks in the shared mini-SWE-agent v2.0.0 harness, in a run dated February 17, 2026, level with GPT 5.2 in SWE-bench Verified, Pro and What Replaced Them: By the Numbers.⁶ In Most-Used AI Models on OpenRouter by Weekly Tokens, GLM 5.3 Flash is second with 12.67 trillion tokens in the seven days to October 12, 2026, GLM 5.3 tenth with 3.96 trillion and GLM 5.2 twentieth with 1.53 trillion.⁷ Z.ai lists GLM-5.3-Flash at $0.50 per 1M output tokens, tied for second in Cheapest LLM APIs per 1M Output Tokens, and GLM-5.3 at $4.40.⁸ GLM-5.3-Flash is published under the MIT License and logged 6,511,401 Hugging Face downloads in the 30 days to October 13, 2026, as covered in Chinese Open-Weight Models: Downloads, Derivatives and Token Share; GLM-5.3 uses the GLM-5.3 License.⁹