Skip to main content
The openai provider supports the OpenAI /chat endpoint with OpenAI-specific default configuration options.
For Azure OpenAI, use the azure provider instead.For other OpenAI-compatible providers (Groq, Ollama, OpenRouter, Together AI, etc.), use openai-compatible.

​
Quick Start

client<llm> MyClient {
  provider "openai"
  options {
    api_key env.MY_OPENAI_KEY
    model "gpt-5-mini"
    temperature 0.1
  }
}

​
Authentication

Set your OpenAI API key as an environment variable: 
export OPENAI_API_KEY="your-api-key-here"
Or specify it explicitly in your BAML configuration: 
client<llm> MyClient {
  provider "openai"
  options {
    api_key env.MY_OPENAI_KEY
    model "gpt-5-mini"
  }
}

​
Configuration Options

​
BAML-Specific Options

These options modify the API request sent to OpenAI.
​
api_key
string
default:"env.OPENAI_API_KEY"
Used to build the Authorization header: Authorization: Bearer $api_key
​
base_url
string
default:"https://api.openai.com/v1"
The base URL for the OpenAI API. Override this to use different regions or custom endpoints.Example for EU region:
client<llm> EUClient {
  provider "openai"
  options {
    base_url "https://eu.api.openai.com/v1"
    model "gpt-5-mini"
  }
}
​
headers
object
Additional headers to send with requests.
client<llm> MyClient {
  provider "openai"
  options {
    model "gpt-5-mini"
    headers {
      "X-My-Header" "my-value"
    }
  }
}

​
Supported Models

​
model
string
required
The OpenAI model to use.
ModelUse CaseContextKey Features
gpt-5Coding, agentic tasks, expert reasoning400K totalBuilt-in reasoning, 45% fewer errors
gpt-5-miniWell-defined tasks, cost-efficient400K totalFaster alternative to GPT-5
gpt-5-nanoLightweight tasks, maximum efficiency400K totalMost cost-effective GPT-5 variant
gpt-4.1Large-scale technical work1MEnhanced coding, instruction following
gpt-4.1-miniBalanced performance and cost1MReplaces GPT-4o mini
gpt-4.1-nanoLightweight variant1MBudget-friendly option
gpt-4oGeneral purpose, multimodal200KUpdated knowledge cutoff June 2024
You can specify any model name - BAML won’t validate whether it exists.

​
Model Parameters

These parameters are passed directly to the OpenAI API.
For reasoning models like o1 or o1-mini, use max_completion_tokens instead of max_tokens. Set max_tokens to null.See the OpenAI API documentation for details.
client<llm> OpenAIo1 {
  provider "openai"
  options {
    model "o1-mini"
    max_tokens null
    max_completion_tokens 4096
  }
}
Common parameters:
  • temperature - Controls randomness (0-2)
  • max_tokens - Maximum tokens to generate
  • top_p - Nucleus sampling parameter
  • frequency_penalty - Reduces repetition (-2.0 to 2.0)
  • presence_penalty - Encourages new topics (-2.0 to 2.0)
For all options, see the OpenAI API documentation.

​
Regional Endpoints

To access OpenAI’s API in different regions: 
client<llm> EUClient {
  provider "openai"
  options {
    base_url "https://eu.api.openai.com/v1"
    model "gpt-5-mini"
  }
}

​
Features

  • Streaming: Supported for real-time response generation
  • Multimodal: Supports text and image inputs
  • Function Calling: Native support for tool use and function calling
  • JSON Mode: Structured output support

​
Do Not Set

​
messages
DO NOT USE
BAML automatically constructs this from your prompt.
​
stream
DO NOT USE
BAML automatically sets this based on how you call the client in your code.

Build docs developers (and LLMs) love

Get started for freeTalk to us