openai-patterns

When to Use

Apply when building or reviewing code that calls the OpenAI API. Covers GPT-4, GPT-4o, o1, and Responses API.

Core Rules

• Always handle rate limits with exponential backoff — they happen in production
• Set max_tokens on every call — unbounded completions cause cost surprises
• Log token usage per call — you need this for cost attribution and anomaly detection
• Use structured outputs (response_format) when you need parseable data — regex on free text is fragile
• Never hardcode API keys — use env vars, rotate quarterly

Model Selection

Model	Best for	Approx cost
gpt-4o-mini	Classification, extraction, simple Q&A	~$0.15/1M input
gpt-4o	Complex reasoning, code generation, long context	~$2.50/1M input
o1-mini	Math, coding problems requiring multi-step reasoning	~$1.10/1M input
o1	Hardest reasoning tasks, research-grade	~$15/1M input

Rule: start with gpt-4o-mini and only upgrade if quality is insufficient. The gap is smaller than you think.

Basic Call Pattern

from openai import OpenAI
import os

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain transformer attention in one paragraph."},
    ],
    max_tokens=500,
    temperature=0.7,
)

text = response.choices[0].message.content
usage = response.usage  # prompt_tokens, completion_tokens, total_tokens

Structured Outputs

from pydantic import BaseModel

class SentimentResult(BaseModel):
    sentiment: Literal["positive", "negative", "neutral"]
    confidence: float
    reason: str

response = client.beta.chat.completions.parse(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": f"Analyze: {text}"}],
    response_format=SentimentResult,
)

result: SentimentResult = response.choices[0].message.parsed
# result.sentiment, result.confidence, result.reason — all typed