OpenAI API

The OpenAI API has been the de facto standard for LLM-powered development since GPT-3 launched in 2020. Developers send messages and receive model completions, with support for tool calling (the model can request execution of functions), structured outputs (JSON-constrained responses), streaming, vision inputs, and audio. The API exposes all major OpenAI models: GPT-5.x variants, GPT-5.3 Codex, image generation via GPT Image, and audio via the Realtime API.

In 2025, OpenAI introduced the Responses API as a replacement for the Chat Completions API. The Responses API adds stateful session management (the API can maintain context across requests without the developer manually managing message history), built-in file search and web search as native tools, and a more ergonomic interface for agentic workflows. The older Chat Completions API still works and will not be removed immediately, but new projects should use Responses. The 'OpenAI-compatible API' format has also become an informal standard: many competing providers (Groq, Together AI, Anthropic via third-party adapters) accept the same request format.

For builders, the OpenAI API is usually the first integration point for AI features because of its broad documentation, large community, and ecosystem of libraries. The pricing model is usage-based (per million input and output tokens), with batch processing discounts of 50%. API access is separate from ChatGPT subscription tiers; a ChatGPT Pro subscription does not grant unlimited API access, and API usage is billed independently.