OpenAI's API surface: the reference against which all others are measured. GPT-4o (multimodal), o1/o3 (reasoning), the Assistants API, structured outputs, function calling, image generation, embeddings, TTS/STT. The most comprehensive product surface in the category.
- 01Name the OpenAI model tiers and their positioning.
- 02State when to reach for chat completions vs. Assistants API.
- 03Explain OpenAI's structured-output guarantee.
The surface
chat.completions (the workhorse), responses API (2025+ recommended path for stateful multi-turn with tools), embeddings, images, audio. Structured outputs via response_format = json_schema are guaranteed schema-conformant — a hard guarantee, not a hope.
Model choice
GPT-4o: strong general-purpose, multimodal. o3/o3-mini: reasoning, expensive per token, dramatically better on math/code/logic. GPT-4o mini: cheap and fast. Choose based on task: reasoning for hard reasoning, GPT-4o for most work, mini for volume.
- OpenAI's structured outputs are guaranteed, not encouraged.
- Reasoning models cost 5-20x per token but earn it on hard tasks.