Pelican on a Bicycle HTML Test — GPT-6 Astra, Gemini 3.8 Flash, and DeepSeek 4.1 Flash (September 2026)

The Test and the Prompt

I gave three AI models the same prompt and asked each to create an HTML page featuring a pelican riding a bicycle.

Here is the prompt:

Create a single HTML file that generates a 2D SVG animation of a pelican riding a bicycle and displays it on the page.

You can try the pages generated by GPT-6 Astra, Gemini 3.8 Flash, and DeepSeek 4.1 Flash here:

• GPT-6 Astra demo
• Gemini 3.8 Flash demo
• DeepSeek 4.1 Flash demo

Test Setup

• VSCode Codex — GPT-6 Astra (Low)
• VSCode Antigravity — Gemini 3.8 Flash (Low)
• TraeCode CN — DeepSeek 4.1 Flash

Comparing the Results

GPT-6 Astra

The scene feels cohesive, with an elegant color palette. The pelican appears to travel from left to right, and the overall motion is consistent with that direction. Its feet also follow a convincing pedaling motion.

The page includes controls to pause and resume the animation and adjust its speed.

There are still some rendering issues where parts of the pelican and bicycle overlap. The layering is not entirely correct.

Gemini 3.8 Flash

This version uses brighter, more vivid colors. The pelican appears to travel from left to right, and the overall motion looks consistent with that direction.

However, its wings do not grip the handlebars, and its feet are not on the pedals.

The page includes pause and resume controls, a speed adjustment, and a button that plays a bicycle bell sound.

DeepSeek 4.1 Flash

The scene is relatively simple, with bright colors. Although the pelican appears to travel from left to right, the road’s movement and the wheels’ rotation are noticeably out of sync.

There is a visible gap in the middle of the pelican’s throat pouch, and its wings are not on the handlebars. Both feet, however, are positioned on the pedals.

The page has no additional controls or features.

Side-by-Side Comparison

Aspect	GPT-6 Astra (Low)	Gemini 3.8 Flash (Low)	DeepSeek 4.1 Flash
Visual design and colors	Elegant, cohesive, and refined	Bright and vivid	Simple and basic
Anatomy and bicycle mechanics	Coordinated pedaling, with minor layering issues	Feet off the pedals; wings not gripping the handlebars	Feet on the pedals, but a visible gap or overlap issue in the throat pouch
Motion and physical consistency	Natural, coordinated motion in the correct direction	Generally coordinated	Road movement and wheel rotation do not align
Interactive features	Play/pause and speed controls	Play/pause, speed controls, and a bicycle bell	No interactive controls

Overall Impressions

Gemini added a playful extra with its bicycle bell button. GPT produced the most polished visual design and animation details, while DeepSeek delivered a simpler interpretation of the basic request.

In this test, using Low reasoning settings or Flash models, the results showed differences in how well the models handled spatial relationships: the rider’s posture, the stacking order of visual elements, and relative motion. Coordinating articulated 2D animation across multiple layers—such as keeping the feet attached to rotating pedals—remains a challenging task for these configurations. GPT-6 produced the smoothest pedaling motion, while DeepSeek showed a directional mismatch in the relative motion of the bicycle, wheels, and ground.

Overall, GPT-6 Astra produced the best result in this comparison.