Skip to main content

OpenAI Responses

The openai-responses provider supports OpenAI's /responses endpoint, which uses the newer Responses API instead of Chat Completions.

tip

OpenAI recommends the Responses API for new users. See OpenAI's comparison guide.

warning

o1-mini is not supported with the openai-responses provider.

Basic Example​

client<llm> MyResponsesClient {
    provider "openai-responses"
    options {
        api_key env.MY_OPENAI_KEY
        model "gpt-4o"
        reasoning {
            effort "medium"
        }
    }
}

BAML-Specific Options​

OptionTypeDefaultDescription
api_keystringenv.OPENAI_API_KEYAPI key
base_urlstringhttps://api.openai.com/v1Base URL
headersobject-Additional headers
client_response_typestring"openai-responses"Response format

Provider Parameters​

reasoning.effort​

Controls the reasoning effort level:

ValueDescription
lowMinimal reasoning effort
mediumBalanced reasoning effort
highMaximum reasoning effort
client<llm> HighReasoningClient {
    provider openai-responses
    options {
        model "o4-mini"
        reasoning {
            effort "high"
        }
    }
}

tools​

Tools the model can use during reasoning:

client<llm> WebSearchClient {
    provider openai-responses
    options {
        model "gpt-4o"
        tools [
            {
                type "web_search_preview"
            }
        ]
    }
}

Available Models​

ModelUse CaseContextFeatures
gpt-4oGeneral purpose128KMultimodal, balanced
gpt-4o-miniCost-efficient128KFast, affordable
o4-miniFast reasoningStandard92.7% AIME, cost-efficient

Use Cases​

Image Input​

client<llm> VisionClient {
    provider openai-responses
    options {
        model "gpt-4o"
    }
}

function AnalyzeImage(img: image) -> string {
    client VisionClient
    prompt #"
        {{ _.role("user") }}
        What is in this image?
        {{ img }}
    "#
}

Advanced Reasoning​

client<llm> AdvancedReasoningClient {
    provider openai-responses
    options {
        model "o4-mini"
        reasoning {
            effort "high"
        }
    }
}

function SolveComplexProblem(problem: string) -> string {
    client AdvancedReasoningClient
    prompt #"
        {{ _.role("user") }}
        Solve this step by step: {{ problem }}
    "#
}

See Also​