Models: OpenAI

See also

• Guide → Providers: guide/providers.md
• Architecture → Provider Abstraction: architecture/provider-abstraction.md#provider-mapping

alloy.models.openai

OpenAIBackend

Bases: ModelBackend

OpenAI backend using the Responses API.

Implements completion and streaming via responses.create/responses.stream, supports function tool-calls by looping with previous_response_id, and emits structured outputs using text.format with a JSON Schema when an output_schema is provided. Raises ConfigurationError if the SDK is unavailable.

Usage

export OPENAI_API_KEY=...
export ALLOY_MODEL=gpt-5-mini

Note:

• Some reasoning models (e.g., gpt-5, o1, o3) ignore or reject temperature settings. Alloy detects these and does not send temperature for such models; a debug log line is emitted when dropped.