OpenAI API Platform

OpenAI API Platform is OpenAI's developer platform for building apps and agents on GPT, image, voice and video models, billed per token with no seat plans.

OpenAI API Platform is the developer platform for building software on OpenAI's language, reasoning, voice, image and video models, sold as metered API access to engineering teams. OpenAI makes and owns it outright. The platform runs on its own subdomain with its own console, billing and documentation, separate from ChatGPT, the chat product built on the same models. The problem it solves is putting a frontier model inside an application without training or hosting one: a team creates a key, calls a model and pays for the tokens it uses. The core is the Responses API, which serves GPT-6 Astra and the GPT-5.6 family with function calling, Structured Outputs, a reasoning effort setting, prompt caching and a Batch API for offline work. Around it sit the specialist models: gpt-5.3-codex for coding agents, gpt-image-2.5-sunburst and gpt-image-2.5-flare for image generation, Sora 2 for video, the Realtime API for speech, and the transcription and embeddings endpoints. AgentKit is the agent layer. Agent Builder lays out multi-step workflows on a visual canvas and the Agents SDK defines the same agents in code. Hosted tools cover web search, file search over vector stores and remote MCP servers. Guardrails, approval steps, tracing and evals watch what an agent does, and ChatKit embeds a finished agent in a website or app. Fine-tuning, Zero Data Retention for eligible customers, project-level data residency and a published deprecation schedule cover the operating side. It fits developers and product teams who ship features on top of a model rather than a finished assistant. That means backend engineers wiring a model into an existing service, startups building an agent product, and data teams pushing large jobs through the Batch API. Enterprises that need residency and retention terms in the contract are served by the same console. Solo developers start there too, since the usage tier grows with spend. A company that wants an assistant for its staff buys ChatGPT instead. It is billed on usage rather than seats: per token for models, per call for hosted tools and per minute for live audio, with a Scale Tier for committed traffic. Each headline model carries its own published per-token rate, with the Batch API and cached input discounted. In practice it sits behind customer-facing assistants, internal copilots, document pipelines and coding tools. Typical jobs are extracting structured data against a schema, answering questions over a company's own files, routing support conversations between specialist agents, and generating images and speech. It connects to a team's own systems through function calling and to third-party tools through MCP connectors. The Agents SDK gives code-first teams the same building blocks as the canvas, so an agent prototyped in Agent Builder moves into a codebase without a rebuild.