A model that handles more than one type of input or output, such as text and images together. It is why you can now show a model a photograph and ask questions about it, or hand it a chart and ask what it shows.
Technical view
Typically achieved by encoding each modality into a shared representation space so that a single model can attend across them. Images are usually split into patches and treated much like tokens. Adding modalities does not automatically confer competence in them: performance on tasks such as reading dense diagrams or precise spatial reasoning remains substantially weaker than on text.
Related
Part of AI Terminology Index, section: Extending and Applying Models.