In actual testing, GLM-4.7's code capabilities surpass those of Claude Sonnet 4.5, making it the open-source state-of-the-art (SOTA) solution.

GLM-4.7 has been released for less than two days, and real-world testing has already flooded the X platform.

Someone used a segmentSimpleThe game screen recording was used to generate a complete, playable mini-game with sound effects in just a few minutes. The entire process generated approximately 6600 tokens at a stable speed of 16 tok/s, and the mini-game was generated in less than 7 minutes!

Not just mini-games. In tool calls,Agent In more complex scenarios such as missions, GLM-4.7 also performs very well: a single dialogue is enough to get all tool calls right.

These feedbacks clearly show that GLM-4.7 not only has stronger generation capabilities, but also...It's easier to complete the entire task in one go, and there's rarely a need for manual intervention..

GLM-4.7 has been significantly enhanced. Coding,Agent and Reasoning Its performance on LiveCodeBench V6 reached 84.9 points, surpassing Sonnet 4.5 in coding ability.up to dateopen source SOTAIt scored 42.8 on HLE, 38% higher than GLM-4.6, exceeding... GPT-5.1.

In today's article, we'll also conduct practical tests and experiences from multiple perspectives.

Currently, GLM-4.7 is available on platforms such as Z.ai and BigModel. You can experience it by selecting GLM-4.7 in the model selector in the upper left corner.

Website:https://chat.z.ai

Case 1: Generating a PPT in one sentence

As the year draws to a close, this is also the time when the demand for PowerPoint presentations is at its highest. Let's try making one:

Select below the input box AI PPT function, enter the followingPrompt words:

Introducing the Forbidden City.

GLM-4.7 first conducted a multi-dimensional search to understand the basic information, historical background, architectural features and main attractions of the Forbidden City.