Z.ai's GLM-5.1: An AI Model That Exceeds Expectations

A Colossal and Creative Model

The Chinese AI lab Z.ai recently unveiled its latest model, GLM-5.1, which impresses with its 754 billion parameters and a size of 1.51 TB on the Hugging Face platform. This model, licensed under MIT, maintains the same size as its predecessor, GLM-5, and shares the same research paper.

GLM-5.1 is accessible via OpenRouter, and during a test, the model was asked to create an SVG representing a pelican on a bicycle. To everyone's surprise, the model not only generated the SVG but also produced an HTML page incorporating CSS animations.

An Unexpected Animation

The SVG generated by GLM-5.1 was of high quality, but the CSS animation encountered issues, placing the pelican off-screen. This anomaly was reported to the model, which explained that the CSS transformations interfered with the SVG positioning. To resolve this, GLM-5.1 suggested separating the SVG positioning from the animation and using <animateTransform> for the rotations, thus ensuring a correct rendering.

Fascinating Animation Details

The model then generated new HTML code that corrected the initial problem. The details of the animation, particularly the oscillation of the pelican's beak, were especially appreciated. Comments in the code accurately describe this animation, highlighting the attention to detail exhibited by GLM-5.1.

Limitless Creativity

Following this demonstration, another suggestion was made on Bluesky to create an "OPOSSUM DE VIRGINIE ON AN E-SCOOTER." The model once again surprised with the richness of details in the HTML+SVG comments, including elements such as "earring sparkles" and a "gradient of opossum fur."

Z.ai's GLM-5.1 does not merely generate images; it pushes the boundaries of AI creativity by integrating complex animations and correcting its own mistakes, opening new perspectives for the creative applications of artificial intelligence.