Pick by task type, context window and price.
The Models page groups by chat, image, video and embeddings. Calling the wrong endpoint for a model type is rejected outright — e.g. a chat model on an image endpoint.
Each model page lists its context window and max output. For long documents, mind the ceiling: exceeding it gets truncated upstream or errors.
Prices vary widely within the same capability tier. For structured work like classification or extraction, a small model is often enough and an order of magnitude cheaper.