ChatGPT API: Complete Guide

The OpenAI API provides access to the GPT-4o, GPT-4o-mini and o1 models. It is the foundation for AI chatbots, content generation and task automation.

Getting an API Key

Sign up at platform.openai.com, top up your balance ($5+) and create an API key. Keep the key in environment variables.

First Request (cURL)

curl https://api.openai.com/v1/chat/completions   -H "Content-Type: application/json"   -H "Authorization: Bearer $OPENAI_API_KEY"   -d '{
    "model": "gpt-4o-mini",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is VPS hosting?"}
    ],
    "max_tokens": 500
  }'

PHP Integration

composer require openai-php/client

<?php
require 'vendor/autoload.php';
$client = OpenAI::client($_ENV['OPENAI_API_KEY']);
$response = $client->chat()->create([
    'model'    => 'gpt-4o-mini',
    'messages' => [
        ['role' => 'system', 'content' => 'You are a ZevsHost support assistant.'],
        ['role' => 'user',   'content' => 'How do I add a domain in cPanel?'],
    ],
    'max_tokens' => 800,
]);
echo $response->choices[0]->message->content;

Python Integration

pip install openai

from openai import OpenAI
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Explain VPS in 3 sentences."}],
    max_tokens=200
)
print(response.choices[0].message.content)

Streaming (streamed output)

stream = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Write an article about VPS"}],
    stream=True
)
for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Pricing and Models (2025)

Model	Input (1M)	Output (1M)	Use case
gpt-4o-mini	$0.15	$0.60	Chatbots, support
gpt-4o	$2.50	$10.00	Complex tasks
o1-mini	$1.10	$4.40	Logic, code

Savings: gpt-4o-mini is 15x cheaper than GPT-4o with comparable quality for simple tasks. Use max_tokens to control costs.

Common API Errors

OpenAI API error codes are standard HTTP codes, but the causes behind them are not always obvious from the response text.

401 — Incorrect API Key Provided

{"error": {"message": "Incorrect API key provided: sk-***", "type": "invalid_request_error"}}

The key failed validation: it was revoked, copied incompletely, or belongs to a different project within the organization. Check the key on the API keys page at platform.openai.com — revoked keys are clearly marked there.

429 — Rate Limit Reached / insufficient_quota

Error 429 actually covers two different problems: exceeding the requests-per-minute limit (temporary, retry with a delay) and running out of paid balance (insufficient_quota, requires topping up). The type field in the JSON response distinguishes between the two cases — they should not be handled with the same retry logic.

Main Error Codes and How to React

Code	Error type	What to do
401	invalid_request_error / bad key	Check and reissue the key
429	rate_limit_exceeded	Retry with exponential backoff
429	insufficient_quota	Top up the balance, retrying will not help
500/503	server_error	Retry later, the problem is on OpenAI's side

When a Direct OpenAI API Call Is Not a Good Fit

User data cannot leave your own infrastructure — in that case a local option such as Ollama is needed.
You need independence from a single provider — it is worth building in an abstraction layer from the start, as described in the article comparing AI APIs.
The budget is unpredictable under peak load — without a hard limit on max_tokens and cost monitoring, the bill can grow faster than expected.
You need a guaranteed response time — the public API offers no latency SLA, and under peak load OpenAI's response can take noticeably longer than usual.

Verifying the Integration Works

curl -s https://api.openai.com/v1/models -H "Authorization: Bearer $OPENAI_API_KEY" | head -c 200

If the key works, the request returns JSON with the list of available models. A 401 error at this step means the problem is with the key, before you even start debugging the chat request logic itself — this narrows down which part of the system is failing much faster.

When designing the integration, build in cost logging from the start: the usage field in the API response contains the exact number of input and output tokens used. Storing these values alongside the request identifier lets you quickly find which usage scenario spiked the bill, instead of investigating after the fact from the total amount in the OpenAI dashboard.