Skip to content

Visual Sketchpad

Give multimodal LLMs a visual sketchpad for reasoning

Visit official site

What Visual Sketchpad is

The framework is described as giving multimodal LLMs a visual sketchpad, letting models operate on visual artefacts they draw while planning and reasoning.

What you can do with it

  • Generate intermediate sketches to reason through tasks
  • Use auxiliary lines to solve geometry problems
  • Call expert vision models for visual perception

Application scenarios

  • Help students solve geometry with auxiliary lines
  • Support researchers in scientific computing
  • Help developers understand complex data structures

Pros & cons

✓ What we like

  • Expert vision models can draw bounding boxes or segmentation masks
  • Gains are described on maths and complex visual reasoning tasks

! What to watch out for

  • Drawing steps are tied to the chosen multimodal LLM
  • Expert models are needed for some perception tasks

FAQ

What does the model draw with?

The material describes lines, boxes and marks instead of text-only reasoning steps.

Which maths tasks are listed?

The feature list names geometry, functions, charts and chess.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review