LLM visual benchmark suite

Same prompts. Different models.

We give every model the same standardized visual brief and compare the results. Pick a model to see what it built.