Text Generation and Chat

Text-generation models can serve single-turn generation, multi-turn conversation, reasoning, and tool-assisted workflows. Choose the model by its declared capabilities rather than assuming every text model accepts the same inputs and controls.

Choose a model

Compare the following on the Models page:

• Context length and maximum output.
• Supported input and output modalities.
• Streaming support.
• Tool calling and structured output.
• Reasoning controls and prompt caching.
• Current availability and price.

A text model can accept images or other media without becoming an image-generation model; input modality and output task are separate concepts.

Choose a client protocol

Send the request through a client protocol supported by the selected model and your SDK. Message roles, tool schemas, multimodal content blocks, stream events, and optional parameters follow that protocol.

See Protocols and SDK Compatibility for the current public protocol surfaces. Using one protocol does not restrict the request to models owned by the protocol's original manufacturer.

Build the conversation

Provide only the conversation history needed for the current response. Your application owns persistent conversation state unless a documented API explicitly says otherwise. Validate model-specific limits before attaching media or large tool schemas.

Handle output and usage

Use Synchronous, Streaming, and Asynchronous Requests for delivery semantics. The selected model determines whether a capability is accepted; unsupported capability-specific parameters can be rejected. Final status, usage, and cost are available in request logs.