Compare two models

Open Model comparison and select two different available text models. Enter one shared prompt and each model's maximum output tokens, then confirm the two independent requests before starting. This view compares outputs; it does not assign a quality score or normalize unlike model capabilities.

The server supplies the existing per-request admission estimate cap, currently $0.20 each / $0.40 combined. This is the admission estimate limit, not a final bill or refund guarantee. Model caps, organization access, subscription credit and cumulative Key budgets still apply. If the policy changes before submission, the request is refused and you must refresh and confirm again.

Each side has its own short-lived Key, response state, stop button and exact receipt. One side failing does not cancel the other or auto-retry. Stop both is available. Received partial output remains readable. Finished output and settled usage are separate states; an empty or delayed receipt is unknown, not zero.

This page accepts text only. Speech, Realtime, file-transcription and image-output models stay in their dedicated experiences. It does not reuse Chat history or automatically submit attachments. The prompt and results stay in page memory; copy or download the Markdown comparison before leaving. Exports contain the original input, model IDs, output token settings, states, partial text and receipt references, without Keys or media.

Changing input or configuration does not run a model. Every comparison needs another explicit usage confirmation. Clipboard failures offer download as an alternative. Actual provider quality, latency and device behavior require separate real-world acceptance.