Z.ai announced GLM-4.6, the latest model in its GLM series, highlighting upgrades across real-world coding, long-context processing, reasoning, and agentic applications. The release is positioned as a notable step forward for Chinese models. GLM-4.6 is available now on Z.ai Chat and the Z.ai API platform, and the company has also open-sourced the model on Hugging Face under the MIT license. Z.ai describes GLM-4.6 as an Artificial Intelligence model designed to push practical capability while improving efficiency and accessibility.
Core technical enhancements include an expanded context window from 128K to 200K tokens, enabling the model to better manage complex agentic workflows. Z.ai reports stronger coding performance not just on standard benchmarks but also within real-world developer tools such as Claude Code, Cline, and Roo Code. Reasoning has been reinforced and now supports tool use, while agentic and writing capabilities have been refined for smoother framework integration and text generation that aligns more closely with human preferences. Efficiency has also been improved, with average token consumption reduced by more than 30 percent compared with GLM-4.5.
In evaluations, GLM-4.6 is said to match the performance of Claude Sonnet 4 and Claude Sonnet 4.5 across eight general capability benchmarks, including AIME 25, GPQA, and SWE-Bench Verified. To assess practical programming ability, Z.ai ran 74 real-world coding tests inside the Claude Code environment, where GLM-4.6 reportedly surpassed Claude Sonnet 4 and other open-weight models. Z.ai has made the underlying test data publicly available on Hugging Face to support external verification and reproduction, emphasizing transparency around methods and results.
Alongside the model debut, Z.ai is upgrading its GLM Coding Plan to deliver a more competitive developer offering. Existing subscribers will be automatically moved to GLM-4.6 and gain new features such as image recognition, search, and support for more than 10 mainstream coding agents. For higher-volume needs, a GLM Coding Max plan is being introduced, which Z.ai states provides three times the usage of the Claude Max (20x) plan. The company framed GLM-4.6 as building on GLM-4.5, which integrated reasoning, coding, and agentic features, and noted that commercial API usage on its platform grew more than tenfold following that earlier release.
