OpenAI API

The OpenAI API gives you programmatic access to every model OpenAI ships: leading models for general intelligence, advanced models for cost-efficient production work, models for reasoning, plus models for images, Whisper for speech, and the embeddings models for semantic search. It is the most-used commercial AI API by a wide margin, and most production LLM stacks start here. The chat completions and responses endpoints are the workhorses. You send a list of messages with roles (system, user, assistant), optionally with tools or structured output schemas, and stream the response back token by token. Function calling lets the model invoke your code, structured outputs guarantee a JSON shape, and the Responses API stitches multi-turn agent loops together with built-in tool use. Model choice drives most of the cost. A recommended production default model at a certain price per million tokens is available: 1M-token context window, beats another model on every benchmark, and supports cached input at a lower price per million tokens for repeated prefixes. Another model at a different price is the right pick for classification and extraction at scale. The flagship model at a higher price is for complex agent and long-context work; a pro version unlocks the highest-stakes reasoning. For hard reasoning, a model at a certain price now matches another model on price and replaced an older, more expensive model. Another model covers reasoning at a fraction of the cost. Image generation runs through a model which superseded a previous version. Whisper handles transcription; a higher-accuracy option is available. Embeddings power semantic search and RAG. Fine-tuning is supported on older generations; the newest model families are not yet fine-tunable. The Batch API discounts every model by 50% in exchange for a 24-hour SLA, the single biggest cost lever for non-real-time workloads.