OpenAI¶

export OPENAI_API_KEY=sk-...

from promptcanary import LiteLLMProvider

provider = LiteLLMProvider("openai/gpt-5.4", temperature=0.0)

Recommended Models¶

Tier	Model string	Notes
Flagship	`openai/gpt-5.5`	Highest capability, highest cost.
Balanced	`openai/gpt-5.4`	Good default for most canary suites.
Fast / cheap	`openai/gpt-5.4-mini`	Suitable for frequent (hourly/daily) checks.

CLI Usage¶

promptcanary run --provider openai/gpt-5.4 --save-baseline
promptcanary compare --provider openai/gpt-5.4 --fail-on-drift

JSON Mode¶

To test structured output specifically, pass response_format through extra_params:

provider = LiteLLMProvider(
    "openai/gpt-5.4",
    extra_params={"response_format": {"type": "json_object"}},
)

Pair this with JsonValidityProbe and JsonSchemaProbe to catch any regression in structured-output reliability.

Cost Awareness¶

OpenAI's flagship models are the most expensive tier in most canary matrices. See Multi-Provider Matrix for a scheduling strategy that runs expensive models weekly while cheaper or free models run more frequently.