Preview terms

Availability and capacity

The preview runs on a single shared GPU with a fixed compute budget. One inference runs at a time across both models and all users. When the service is busy, wait for the Retry-After interval before trying again.

There is no uptime guarantee or service level agreement. The service can be slow, busy, or unavailable. If the compute budget runs out, the API stops responding until the budget resets. Models, limits, and the preview itself can change or end.

Evaluate the answers

Model outputs are predictions. Check their suitability for your task before relying on them. Use this preview to prototype and evaluate, rather than as production infrastructure.

Data and your own deployment

Use nonconfidential inputs and read the privacy notice. If you need private data handling or guaranteed capacity, contact us on X about deploying on your own compute.

The open source models have their own licenses: Kev and Strands Decider.