Understanding Model Specs

Reading a Model Page

Each model in the Models directory records a consistent set of technical specifications so you can compare options on the same terms.

Context and Output

• Context window: the maximum number of tokens the model can take as input.
• Max output tokens: the upper limit on tokens the model can generate in a single response.

Modalities

The modality field lists what kinds of input or output the model handles, such as text, image, audio, or video. A model can support several at once.

License and Weights

• License names the terms the model is released under.
• Open weights indicates whether the model's weights are publicly available, as opposed to being offered only through an API.

Capabilities

Capability flags tell you what the model can do:

• Tool use (function calling)
• Vision (image understanding)
• Audio input or output
• Video input
• Streaming responses

Other Details

Model pages may also include the release date, the training data cutoff, pricing notes, and links to the official site, model card, documentation, paper, Hugging Face, or GitHub when available.

Need More Help?

To weigh two models against each other, see Comparing Models, or contact support.