How much does the OpenAI API cost?

The short answer: OpenAI's API is pay-as-you-go, billed per million tokens, with no subscription and no per-request fee. What you pay depends almost entirely on two choices — which model you call, and how many output tokens it writes. The cheapest current model costs a hundred times less per token than the flagship, so “how much does it cost?” is really a question about your workload. Every price below is from OpenAI's official pricing page, checked 6 November 2026.

OpenAI API prices per million tokens (Standard tier)

OpenAI lists separate rates for input tokens (what you send), cached input (repeated prompt content it has already seen), and output tokens (what the model writes back). Output is the expensive side — exactly five times the input rate on every model below.

Two footnotes from the same page: gpt-5.6-sol's price is promotional and holds only until 30 November 2026, so budget for a possible rise after that date. And regional processing (data residency) endpoints add a 10% uplift on models released on or after 5 March 2026.

What one request actually costs

Take a typical chat request: 2,000 input tokens in, 500 output tokens out. On gpt-6.1-terra that is (2,000 ÷ 1,000,000 × $1.00) + (500 ÷ 1,000,000 × $5.00) = $0.002 + $0.0025 = $0.0045 per request. The same request on the flagship gpt-6-astra costs $0.02 + $0.025 = $0.045 — ten times more for the same tokens. Paste your own prompt into the token counter to see where your requests actually land before you pick a model.

What a month costs at real volumes

Scale the same maths to a month of 10 million input tokens and 2 million output tokens:

• gpt-6.1-terra: $10.00 input + $10.00 output = $20/month
• gpt-6.1-sol: $20.00 input + $20.00 output = $40/month
• gpt-6-astra: $100.00 input + $100.00 output = $200/month

Same workload, same token counts — a 10× spread decided by the model dropdown alone. Run your own volumes through the monthly cost estimator and compare models side by side on the model price comparison before you commit a feature to one model.

Three things that move your bill more than the model

1. Output tokens cost five times input tokens. A model that writes long, rambling answers costs far more than one that writes short, direct ones — even at the same input price. Short system prompts and tight output instructions are the cheapest optimisation you will ever make.

2. Cached input costs a tenth or less. Repeated prompt content — system prompts, few-shot examples, retrieved documents — bills at $0.05 per million on terra instead of $1.00. If your prompts repeat the same large prefix every call, caching is worth more than switching models.

3. Tools bill separately. Built-in tools are charged on top of tokens: web search is $10.00 per 1,000 calls and file search tool calls are $2.50 per 1,000 calls. An agent that searches the web on every turn can quietly cost more in tool calls than in tokens.

Bottom line

For most builders the OpenAI API costs somewhere between a few dollars and a few hundred dollars a month — and the number is set by model choice, output length and caching, not by any subscription. Price your workload on the cheapest model that clears your quality bar, count the tokens before you ship, and re-check the official page when your bill moves: OpenAI changes prices, which is why the prices in our calculator are editable and date-stamped. If your spend swings month to month, the quarterly API cost review shows how to catch the drift early — and because usage-based spend moves your taxable profit too, your tax reserve should vary with your API bill.

Price note. All prices above are Standard-tier rates per 1M tokens from OpenAI's official pricing page, platform.openai.com/docs/pricing, checked 6 November 2026. Prices change — verify on the official page before budgeting, and treat the calculator's editable prices as the live source for your own estimate.