Skip to content

OpenAI

Setup

export OPENAI_API_KEY="sk-..."

No extra install — included with pip install llmgate.

Model prefixes

Models starting with gpt-, o1-, o3-, or chatgpt- are routed to OpenAI automatically.

completion("gpt-4o", messages)
completion("gpt-4o-mini", messages)
completion("o3-mini", messages)
completion("o1-preview", messages)

Vision

OpenAI supports both URL and base64 images with an optional detail hint:

completion("gpt-4o", messages=[{
    "role": "user",
    "content": [
        {"type": "text", "text": "Describe this image."},
        {"type": "image_url", "image_url": {"url": "https://...", "detail": "high"}},
    ],
}])

detail values: "auto" (default) · "low" (85 tokens) · "high" (full resolution, more tokens)

Structured outputs

Uses native json_schema with strict: true — the most reliable structured output mode available:

from llmgate import parse

movie = parse("gpt-4o-mini", messages, response_format=Movie)

Notes

  • o1-* and o3-* reasoning models do not support system messages or temperature. Pass as extra_kwargs or omit.
  • frequency_penalty and presence_penalty are passed through via **extra_kwargs.