OpenAI API: Models and Endpoints Overview
You'll learn: How OpenAI organizes its GPT models and the API endpoints you use to interact with them for text generation.
The Core Idea
OpenAI provides access to its GPT models through a web API—you send requests over the internet instead of running models locally. Think of it like ordering food delivery: you don't need the kitchen (GPU hardware) at home, you just send your order (API request) and receive the result.
Available Models
OpenAI offers several GPT model families:
- GPT-4 models (e.g.,
gpt-4,gpt-4-turbo): Most capable, higher cost - GPT-3.5 models (e.g.,
gpt-3.5-turbo): Faster, more affordable - Each model has different pricing, speed, and capability trade-offs
Main Endpoints
Chat Completions (/v1/chat/completions): The modern, recommended endpoint. You send a conversation history (messages with roles like "user" and "assistant"), and the model continues the conversation.
Completions (/v1/completions): Legacy endpoint for simple text continuation. You provide a prompt, the model completes it. Mostly replaced by chat completions.
Key Parameters
When making requests, you control behavior through parameters:
model: Which GPT version to usetemperature: Randomness (0 = deterministic, higher = more creative)max_tokens: Maximum length of the responsemessages(chat) orprompt(completion): Your input text
Key Takeaway: OpenAI's API abstracts away infrastructure complexity—you choose a model and endpoint, configure generation parameters, and let OpenAI handle the hardware and optimization. This API-first approach prioritizes convenience over control, perfect for rapid development.