GPT AI

GPT (Generative Pre-trained Transformer) is OpenAI's large language model family; ChatGPT is the consumer/team chat product built on it. "GPT" is also used generically in the industry for the transformer-based autoregressive LLM architecture GPT popularized, distinct from OpenAI's specific named models.

Two API shapes

OpenAI has two overlapping ways to call a model:

• Chat Completions API (/v1/chat/completions) — the original, widely-copied shape: a messages array of {role, content}, a model, and completion parameters. Most third-party tools and "OpenAI-compatible" APIs from other vendors mimic this shape specifically.
• Responses API (/v1/responses) — OpenAI's newer, more agent-oriented API: built-in support for multi-turn state via a previous_response_id, built-in tools (web search, file search, code interpreter), and a unified way to mix text/image/tool-call items in one response.

from openai import OpenAI
client = OpenAI()

# Chat Completions
resp = client.chat.completions.create(
    model="gpt-4o",  # verify current model id against docs
    messages=[{"role": "user", "content": "Say hi in one word."}],
)
print(resp.choices[0].message.content)

# Responses API
resp = client.responses.create(
    model="gpt-4o",
    input="Say hi in one word.",
)
print(resp.output_text)

Function calling / tools and structured outputs

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]
resp = client.chat.completions.create(
    model="gpt-4o", messages=[...], tools=tools,
)

• Function calling — declare a JSON-schema function; the model returns a structured call for the caller to execute and feed the result back in a follow-up message.
• Structured outputs (response_format={"type": "json_schema", ...}) — constrains the model's raw output to conform exactly to a given JSON schema, rather than hoping a prompted "respond in JSON" instruction is followed.

Model family shape

OpenAI ships multiple concurrent model lines that trade off capability, latency, and cost (historically including a flagship multimodal line and smaller/faster variants, plus separate reasoning-focused model lines). Exact current model names, context windows, and pricing change frequently — don't hard-code a specific id or number from training data into anything user-facing; check OpenAI's model list.

Common pitfalls

• Assuming Chat Completions and Responses are interchangeable drop-ins — they have different request/response shapes and different built-in tool support; migrating between them is a real porting task, not a rename.
• Treating "OpenAI-compatible API" (offered by many other vendors/local runners) as guaranteeing full feature parity — most only implement a subset of the Chat Completions shape (basic messages + streaming), not every parameter or the Responses API.
• Hard-coding a model name in production code without a fallback plan — OpenAI deprecates older models on a published schedule; watch for deprecation notices rather than discovering it at a hard cutover.
• Confusing "GPT" the architecture family with "OpenAI's GPT models" specifically — many other vendors' models (including open-weight ones) are also GPT-style transformers; "GPT" alone doesn't imply OpenAI.

Learn more

• OpenAI API docs
• OpenAI API reference
• Model deprecations