The Chinese-made AI model 'GLM-5.1' has been released for free, beating Claude Opus 4.6 in some tests and demonstrating its power in agent tasks lasting several hours.



Z.ai, an AI company based in China, released its AI model ' GLM-5.1 ' on April 7, 2026. GLM-5.1 boasts world-class performance as an open model, and in some benchmark tests, it has recorded scores that surpass closed models such as GPT-5.4 and Claude Opus 4.6.

GLM-5.1: Towards Long-Horizon Tasks

https://z.ai/blog/glm-5.1

GLM-5.1 - Overview - Z.AI DEVELOPER DOCUMENT
https://docs.z.ai/guides/llm/glm-5.1




The graph below shows the average scores from multiple coding performance benchmark tests. The GLM-5.1 not only outperformed open models like the MiniMax M2.7 and Kimi K2.5, but also recorded a higher score than the closed model Gemini 3.1 Pro.



The benchmark results are as follows. GLM-5.1 outperforms cutting-edge models such as GPT-5.4 and Claude Opus 4.6 in multiple tests.



Existing AI models had a weakness: even after working on a task for a long time, the results would plateau after the first few attempts. GLM-5.1 is designed to be effective in long-duration agent tasks, and the quality of the deliverables can be improved with each additional attempt.

The graph below shows the number of trials (horizontal axis) and the database processing speed (vertical axis) when performing the task of 'designing a high-speed database management system' using GLM-5.1. GLM-5.1 succeeded in gradually improving the system's performance through more than 600 trials.



The following are the results of

KernelBench , which measures the performance of the task 'Optimizing the GPU kernel.' While the kernel performance of GLM-5 (red) and Claude Opus 4.5 (yellow) plateaued after around 200 trials, GLM-5.1 (blue) was able to consistently improve performance over 1200 trials. However, Claude Opus 4.6 (green) performed better in this test.



The following is the result of instructing GLM-5.1 to create a 'Linux desktop-style web application containing 50 apps.' GLM-5.1 succeeded in creating the basic UI framework within the first hour.



The taskbar and window implementation was completed two hours after the output started.



After 4 hours, we successfully implemented a browser, and after 5 hours, an instant messenger. We then completed the task of 'creating 50 apps' in 8 hours.



GLM-5.1 is available at the following link. It is licensed under the MIT License.

zai-org/GLM-5.1 · Hugging Face
https://huggingface.co/zai-org/GLM-5.1



Related Posts:

in AI, Posted by log1o_hf