Ask a brilliant writer for a portrait and you get a vivid description, not a painting. Painting is a different craft.
A language model produces text. Pictures come from a different kind of model, an image model, and a product only “generates images” when such a model is wired in as a tool. ChatGPT has one built in. Claude, as of September 2026, does not. That is a product decision by Anthropic, not a sign that one model is smarter than the other.
What happens if you ask Claude for an image
It tells you it cannot generate a picture and offers what it can do:
- Draw with code. Diagrams, charts, icons and simple illustrations as SVG, HTML, or Mermaid. These are text files that browsers display as graphics, so this works surprisingly well for anything schematic.
- Use a connected image tool. Through a connector (MCP) Claude can call an image model such as GPT Image or Flux and hand you the result. Then it “generates images” like any other product, because it gained the tool.
- Work with the picture you bring. Claude can look at images: describe a screenshot, read a chart, check a design. Not generating is not the same as not seeing.
Example
“Make me a poster for the team offsite.” In Claude: a clear no, then an offer to write the copy and lay out the poster as HTML you can print, or to call an image tool if one is connected. In ChatGPT: a poster, generated by its built-in image model.
Why it matters for you
This is the tools lesson in one picture. The same request succeeds or fails depending on which tools the agent has, not on how clever the model is. When a task needs a capability your agent lacks, connect the tool or use the product that has it, then bring the result back as a file in your folder.
Related
- Tool
- Chat, Work and Code - products differ by their tools
- Model - what a language model can and cannot produce