Inputs, outputs and features
AI model capabilities
What each model takes in, what it puts out, and what the sources say it can do. A blank cell means nothing was listed. It never means no. All models.
2,276 models
✓ yes✕ no, and a source says so nothing listed
What each column can say
The left group has a source that publishes both yes and no, so each column reports three states. Counted over 2,276 models:
- Tools — 149 yes, 0 no, 1 not listed
- Reasoning — 129 yes, 20 no, 1 not listed
- Structured — 149 yes, 0 no, 1 not listed
- Open weights — 75 yes, 74 no, 1 not listed
- Image in — 113 yes, 36 no, 1 not listed
- Audio in — 28 yes, 121 no, 1 not listed
- Video in — 44 yes, 105 no, 1 not listed
- PDF in — 90 yes, 59 no, 1 not listed
- Image out — 20 yes, 129 no, 1 not listed
- Audio out — 2 yes, 147 no, 1 not listed
The right group has no such source. Every value it carries is a yes — there is not one no in the corpus — because the source lists the capabilities a model has and stays silent about the rest. So those columns can be filtered for is listed and never for lacks. 147 of 2,276 models carry at least one listed capability.
Common questions
- What does a blank cell mean?
- Nothing was listed. It does not mean no. The two are different answers and this table keeps them apart: a tick is a yes, a cross is a source that says no, and an empty cell is silence.
- Why can I filter for models that lack tool calling, but not for models that lack web search?
- Tool calling comes from a source that publishes both true and false, so a no is a measurement. The listed-capability columns come from a source that publishes only the capabilities a model has. It never writes a no, so this site cannot offer one.
- Where do the modality columns come from?
- From the input and output modality lists. A list that exists and omits image is a real no, because the source enumerated the modalities and image was not among them. A model with no list at all reads as not listed.
- Is a capability the same as support at every host?
- No. These are model-level facts as the sources describe them. A host can expose a subset. Open the model page to see the hosts.