OpenAI API

by OpenAI

OpenAI's developer API — billed per token, separate from a ChatGPT subscription.

Model APIs & Infra Pay per token; no monthly platform fee

Visit official site → Featured badge ← Back to directory

What is OpenAI API?

The OpenAI API is how software calls GPT, image, audio, and embedding models. It is not included in ChatGPT Plus or Pro.

There is no seat fee; you add a card and pay per million tokens (input, cached input, and output priced separately). Official US rates we checked in September 2026: GPT-6 Astra $10 / $1 cached / $50 output; GPT-5.6 Sol $4 / $0.40 / $20 (Sol was on a promotional rate OpenAI said held at least through 21 November 2026); GPT-5.6 Terra $2 / $0.20 / $12; GPT-5.6 Luna $0.20 / $0.02 / $1.20.

Older GPT-4o mini is still listed at $0.15 / $0.075 / $0.60. Cached input is typically 10% of input.

Batch API halves standard rates. Requests over roughly 270K–272K context tokens bill input at 2× and output at 1.5× on the current tables.

Function calling, structured outputs, vision, image generation, and the Responses API share the same token meter as the model you pick.

A concrete walkthrough: price a support-ticket classifier before you pick Astra

Estimate 50,000 tickets a month, 800 input tokens of thread + 80 output tokens of label JSON. On Luna that is about 40M input + 4M output ≈ $8 + $4.80 = ~$13/month.

On Terra the same volume is about $80 + $48 = ~$128. On Astra it is about $400 + $200 = ~$600 before caching.

If the system prompt is 1,200 tokens and repeats, cached input (10% of input) is the first optimization — not jumping to Sol. Run a week on Luna, score accuracy on a labeled set, and only move the failures to Terra.

Put the overnight backfill on Batch at half price. None of this runs on a ChatGPT Plus seat.

Key features

Same account for chat, reasoning, vision, image, speech, and embeddings — pick a model per call
Function calling and structured outputs for JSON you can parse without regex
Prompt caching (cached input listed at 10% of input on current GPT-5.6 / GPT-6 tables)
Batch API at half the standard token price when the job can wait
No monthly platform fee; ChatGPT Plus/Pro does not include API credits
Official SDKs plus usage dashboards and spend limits in platform.openai.com

How to get started

Create a project at platform.openai.com, add a billing method, and mint an API key — a ChatGPT login is not enough
Start on Luna or a mini/nano model, log tokens for a week, then decide whether Terra/Sol quality is worth the multiplier
Turn on prompt caching for large repeated prefixes (system prompts, retrieved docs)
Send overnight jobs through Batch instead of the realtime endpoint
Set a hard monthly spend limit before you wire the key into production

OpenAI API pricing

No subscription. September 2026 official list, per 1M tokens, short context: GPT-6 Astra $10 in / $50 out; GPT-5.6 Sol $4 / $20; GPT-5.6 Terra $2 / $12; GPT-5.6 Luna $0.20 / $1.20.

Cached input is 10% of input. Batch is 50% off.

Long-context rows cost more. ChatGPT plans do not include this usage.

Plan	Price	Best for
GPT-5.6 Luna	$0.20 / $1.20 per 1M	High-volume extract, tag, cheap chat
GPT-5.6 Terra	$2 / $12 per 1M	Default production quality
GPT-5.6 Sol	$4 / $20 per 1M (promo into late 2026)	Harder reasoning without Astra prices
GPT-6 Astra	$10 / $50 per 1M	Frontier quality; watch the output line
Batch API	50% of the above	Jobs that can wait hours

Input is cheaper than output; cached input is cheaper again. A ChatGPT subscription does not pay this bill.

Our take: Default new features to Luna or Terra. Measure tokens for a week before anyone 'just uses Astra'. If the prompt prefix is stable, caching will move the bill more than arguing about Sol vs Terra. ChatGPT Plus is a different product — do not expect it to cover API invoices.

Prices verified 2026-09. Plans change often — confirm on the official site above before you buy.

Use cases

Product features that need GPT inside your own UI, not chatgpt.com
Batch classification, extraction, or summarization at Luna/mini prices
Agents that call tools via function calling on a stronger model only when the router says so
Image or vision calls on the same key as text, with a separate line on the invoice

Strengths and tradeoffs

Where it stands out

No platform subscription — a prototype can cost a few dollars if you stay on Luna/mini
Caching and Batch are first-class levers, not hidden enterprise add-ons
One key covers text, tools, vision, and images when you actually need that mix

Tradeoffs

Output tokens are several times input; chatty agents get expensive in a way the playground hides
A ChatGPT subscription does not include API usage — teams confuse this every quarter
Model names and list prices move; a spreadsheet from last quarter is already wrong

OpenAI API alternatives

Anthropic API: Claude's API is the usual swap when long documents and careful tool use matter more than OpenAI's multimodal catalog.
Gemini API: Gemini API is often cheaper at the low end and ties into Google Cloud billing if you already live there.
OpenRouter: OpenRouter is one key in front of many vendors, including OpenAI, when you do not want to hard-wire a single provider.
Together AI: Together is the usual home for hosted open-weight models if you are leaving the GPT family.

Stay on OpenAI's API when you want the GPT family plus images/speech on one invoice. Switch to Anthropic for long-document Claude behavior.

Use Gemini API inside Google Cloud. Use OpenRouter when vendor lock-in is the risk you are managing.

FAQ

Is OpenAI API free to use?

OpenAI API's plans: Pay per token; no monthly platform fee. Pricing and free-tier limits change over time, so check the official site above for the latest details.

What is OpenAI API used for?

OpenAI's developer API — billed per token, separate from a ChatGPT subscription.

Who is OpenAI API best for?

OpenAI API is a good fit for product features that need GPT inside your own UI, not chatgpt.com, or batch classification, extraction, or summarization at Luna/mini prices.

What are the alternatives to OpenAI API?

Commonly compared alternatives include Anthropic API, Gemini API, and OpenRouter, see the comparison above for how they differ.

How much does OpenAI API cost?

There is no monthly API fee. You pay per million tokens at the model's list rate. In September 2026 Luna was $0.20/$1.20, Terra $2/$12, Sol $4/$20, Astra $10/$50 (input/output). Batch is half. ChatGPT Plus/Pro does not include API usage.

What's the best OpenAI API alternative?

Anthropic API is the most commonly recommended swap: claude's API is the usual swap when long documents and careful tool use matter more than OpenAI's multimodal catalog. Which one is actually best depends on which of those tradeoffs matters more for what you're doing.

Does ChatGPT Plus include the OpenAI API?

No. ChatGPT is the consumer app. The API is billed on platform.openai.com from a prepaid or invoiced credit balance. Plus, Pro, Business, and Enterprise ChatGPT seats do not pay API invoices.

Which model should I call first?

Start with GPT-5.6 Luna (or a listed mini/nano) for classification and extraction. Move individual routes to Terra or Sol only after you measure quality. Reserve GPT-6 Astra for jobs where the output line is worth $50 per million tokens.

Related tools

Last updated: 2026-09-21 · Reviewed by AIKetra editors · Domain registered 2007 (19 years old) · How we evaluate tools · Embed a Featured badge