Product overview
Think in application responsibilities
Together AI’s documentation centers on running, training and serving open models. It describes serverless inference, dedicated deployment and model customization. For a builder, the decision begins with the application’s responsibilities: conversation, retrieval, structured output and media creation require different tests. This guide focuses on that architectural distinction rather than promising interchangeable services.
01Run a supported model
Serverless inference provides a route to supported models without provisioning your own deployment. Check the selected model’s input, output and billing units. Compatibility at an API surface does not make every model behavior identical.
02Plan customization separately
Together AI documents fine-tuning and deployment of customized models. Dataset preparation, evaluation and serving are separate decisions; a few prompt changes are not equivalent to training, and a hosted model is not automatically transferable elsewhere.
03Distinguish media from language tasks
Together’s current documentation also lists image, video and speech APIs. Evaluate those endpoints on their own terms. A strong text application does not by itself establish product-image fidelity, video continuity or suitability for a finished campaign.