LLM Analog Clock Test

AI, LLMs

Hello everyone, I thought I'd make a post here regarding my number one test question for coding LLMs lately:

"Please generate an html file with css and javascript embedded which implements an analog clock display. It should display the current system time. Make sure to add hour labels around the clock. Also, there should be a digital read-out below the analog clock, and a light and dark mode toggle."

This prompt tests spacial understanding to solve, as well as basic trigonometry. So far I have yet to find a small model able to solve this on my machine locally in one shot. Qwen3-30b a3b comes closest (but required manual correction to work.) I decided to enhance the test for more powerful LLM models (in this case, ChatGPT) with further refinements to the prompt to achieve this result. The user's timezone and time are shown in the upper middle clock but it can be configured per clock to whatever the user wants.

Here's Gemini 2.5 Pro Experimental one shotting it:

Gemini 2.5 Pro Experimental Design

Here's what DeepSeek r1 comes up with:

Deepseek R1_distill_llama_70b

Grok 3 does pretty well with this as well.

Grok 3 clock

Claude 3.7s result doesn't look too great

Claude 3.7 clock