Introduction to the OpenAI API

The OpenAI API provides programmatic access to OpenAI’s models including GPT-4o, o1, and o3. It is the most widely used LLM API and the reference point for how most AI SDKs and frameworks are designed.

Core Endpoints

Chat Completions

The primary endpoint for conversational and instruction-following tasks. Accepts a list of messages with roles (system, user, assistant) and returns a completion.

from openai import OpenAI
client = OpenAI()

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the capital of France?"}
    ]
)
print(response.choices[0].message.content)

Embeddings

Converts text into dense vector representations for semantic search, clustering, and RAG pipelines.

embedding = client.embeddings.create(
    model="text-embedding-3-small",
    input="The quick brown fox"
).data[0].embedding

Structured Outputs

Constrain the model to return output matching a JSON Schema — the most reliable way to get structured data.

response = client.beta.chat.completions.parse(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Extract: John is 30."}],
    response_format=MyPydanticModel
)

Tool / Function Calling

Define functions the model can call. The model returns a structured tool call when it needs external information.

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get weather for a city",
        "parameters": {"type": "object", "properties": {"city": {"type": "string"}}}
    }
}]

Models Overview

Key Parameters

• temperature (0–2): Controls randomness. 0 for deterministic, 1 for default, higher for creative tasks.
• max_tokens: Limits response length.
• top_p: Nucleus sampling — an alternative to temperature.
• stream: Set to True for streaming token-by-token output.

Best Practices

• Always set a system prompt to define the model’s behavior and persona.
• Use structured outputs or function calling whenever you need parseable data.
• Implement retry logic with exponential backoff for rate limit errors.
• Log prompts and responses during development for debugging.
• Monitor token usage — cost scales linearly with tokens.