OpenAI API Pricing: $0.05 to $10 per Million Tokens

In brief

OpenAI API pricing runs from $0.05 per million input tokens on GPT-6 Luna (Flex only) to $10.00 on GPT-6 Astra at Standard, as of 22 November 2026. GPT-6.2 Sol costs $2.00 in and $10.00 out, so 1,000 requests of 2,000 tokens in and 500 out cost $9, or $4.50 through Batch.

How much does the OpenAI API cost per model?

GPT-6.2 Sol costs $2.00 per million input tokens and $10.00 per million output tokens, and GPT-6.2 Astra costs four times that. We read every row on OpenAI's pricing page on 22 November 2026, in US dollars per million tokens, on the short context setting.

A brushed silver OpenAI knot on a black plinth beside two stacks of frosted glass tiles on their own plinths, the right stack twice as tall as the left, in a charcoal studio under white rim light.
Model, read 22 November 2026InputCached inputOutputBatch, in and outFlex, in and outFast, in and out
GPT-6.2 Astra$8.00$0.40$40.00$4.00 and $20.00not listed$16.00 and $80.00
GPT-6.2 Sol$2.00$0.10$10.00$1.00 and $5.00not listed$4.00 and $20.00
GPT-6 Astra$10.00$1.00$50.00$5.00 and $25.00$5.00 and $25.00$20.00 and $100.00
GPT-6.1 Sol$2.00$0.10$10.00$1.00 and $5.00$1.00 and $5.00$4.00 and $20.00
GPT-6.1 Terranot listednot listednot listednot listed$0.50 and $2.50not listed
GPT-6 Lunanot listednot listednot listednot listed$0.05 and $0.25not listed

Cached input is a tenth of the full input rate on every Standard row. The page also prices writing a prompt into the cache: GPT-6.2 Sol charges $2.50 to hold it for 30 minutes and $5.00 for 12 hours.

Terra and Luna deserve a warning. OpenAI's Standard table lists four models and neither is one of them. Their Flex rows carry cached rates of $0.025 and $0.005, but Flex is a beta that OpenAI says is winding down. Our Gemini API pricing page quoted Luna at $0.10 in on 8 November, and the Standard table no longer shows it.

What do Batch, Flex and Fast change?

Batch costs half of Standard, Fast costs twice as much, and Flex matches Batch while it lasts. GPT-6.2 Sol input is $2.00 on Standard, $1.00 on Batch and $4.00 on Fast. Output moves the same way, from $10.00 to $5.00 or $20.00.

OpenAI has added two newer wrinkles. For GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna, eligible background requests are 30% below Standard in a 2 hour window and 40% below in an 8 hour window. On GPT-6.2 Sol that puts input at $1.40 or $1.20 per million, and output at $7.00 or $6.00.

Flex is ending. The page says it stays on the listed models while the beta winds down, no new models will get it, and the retirement date will be published in the Flex processing guide. Don't build a plan on it.

Speed has its own ladder. The page says "Priority processing was renamed Fast mode on July 30, 2026." Ultrafast sits above it, and its price depends on the date:

Ultrafast, per million tokensInputCached inputOutput
GPT-6.2 Astra, 3 times Standard through 28 February 2027$24.00$1.20$120.00
GPT-6.2 Astra, 6 times Standard from 1 March 2027$48.00$2.40$240.00
GPT-6.2 Sol, 3 times Standard through 28 February 2027$6.00$0.30$30.00
GPT-6.2 Sol, 6 times Standard from 1 March 2027$12.00$0.60$60.00

If you are not sure which one you need, start on Standard and move a job to Batch once you know that nobody is waiting on it. Two uplifts apply on top of all of this. Regional processing (data residency) endpoints add 10% for models released on or after 5 March 2026, and FedRAMP endpoints add 10% as well.

What do 1,000 OpenAI API requests cost?

1,000 requests of 2,000 tokens in and 500 out cost $9 on GPT-6.2 Sol and $36 on GPT-6.2 Astra. We worked each figure from OpenAI's rates as we read them on 22 November. Batch halves every one.

Batch cuts 1,000 requests to half of the Standard price
BatchStandard$0$10$20$30$40US dollars per 1,000 requests of 2,000 tokens in and 500 out, OpenAI rates read 22 Nov 2026GPT-6 Astra$22.50$45batch is halfGPT-6.2 Astra$18$36batch is halfGPT-6.2 Sol$4.50$9batch is half

Here's the math for Sol. The input side is 2,000 tokens at $2 per million, or $0.004. The output side is 500 tokens at $10 per million, or $0.005. That makes $0.009 a request, or $9 for a thousand. Astra on the same job is $0.016 in and $0.020 out, so $0.036.

The sum assumes every output token is visible. If a model spends tokens reasoning before it answers, your real output count is higher than the text you see. Count one real call before you trust any estimate, because the visible text is only part of what you pay for.

What does a month of OpenAI API usage cost?

A month of 100,000 requests costs $900 on GPT-6.2 Sol and $3,600 on GPT-6.2 Astra at Standard rates, before tax. That is 200 million tokens in and 50 million out. Here is how the bill moves when you change one setting:

100,000 requests a month, 200M in and 50M outMonthly cost
GPT-6.2 Astra, Standard$3,600
GPT-6.2 Astra, Batch$1,800
GPT-6.2 Sol, Standard$900
GPT-6.2 Sol, Batch$450
GPT-6.2 Sol, 1,500 token prompt cached$615
GPT-6.2 Sol, Fast$1,800
GPT-6.2 Sol, Ultrafast through 28 February 2027$2,700
GPT-6.1 Terra, Flex$225
GPT-6 Luna, Flex$22.50

The cached row assumes a 1,500 token system prompt opens every request. Those tokens bill at $0.10 per million instead of $2.00, which takes the bill from $900 to $615, about 32% off. It leaves out the cache write charge, which is small when requests arrive inside the 30 minute window.

Ultrafast on Sol runs $2,700 now and $5,400 after 1 March 2027, three times Standard and then six. Only pay that where a person is waiting on the answer, and use Batch for everything that can sit in a queue.

What do three real jobs cost on the OpenAI API?

A ticket tagger costs $125 a month on GPT-6.2 Sol, a contract Q and A costs 29 cents for 20 questions, and a search backed answer costs 8.5 cents. These are example jobs written for this page, with token sizes we assumed, priced at OpenAI's rates on 22 November 2026.

Job and assumptionModel and settingCost
Tag 50,000 support tickets, 1,000 tokens in and 50 outGPT-6.2 Sol, Standard$125
Same 50,000 ticketsGPT-6.2 Sol, Batch$62.50
Same 50,000 ticketsGPT-6 Luna, Flex$3.13
Ask 20 questions of one 30,000 token contract, 800 tokens out each, uncachedGPT-6.2 Sol, Standard$1.36
Same 20 questions, contract cached after the firstGPT-6.2 Sol, Standard$0.29
One answer using 3 web searches, 20,000 tokens in, 1,500 outGPT-6.2 Sol, Standard$0.085

The ticket job shows why model choice beats every discount. Batch saves $62.50, and moving to Luna on Flex cuts the bill to $3.13, though Flex is closing and Luna has no Standard row. The contract job shows caching at work: the first call pays the $2.50 write rate on 30,000 tokens, which is $0.075, and the next 19 calls read the contract at $0.10 per million.

The search job has a fixed part. Web search costs $10.00 per 1,000 calls, so three calls are 3 cents, and the retrieved text bills at the model's rates on top.

Is the OpenAI API free?

Not for new accounts. OpenAI's pricing page lists no free usage tier for its models, only 1 GB of free file search storage and 1 GB of free Agent Kit storage. Every model row in the tables above is billed per token from the first call that you make.

One exception, read on OpenAI's API credits guide on 22 November 2026: eligible ChatGPT Pro and Business subscribers can claim monthly API credits. They are $100, $200 or $500 a month on the three Pro tiers, $20 per Standard seat and $100 per Premium seat on Business. Unused credits expire at the end of the billing cycle, and the credits are still subject to API rate limits. Google differs here: its Gemini 3.8 Flash row has a free tier, and Gemini 4 Argon does not. If you only want to chat with GPT, ChatGPT is a separate product from the API, and ChatGPT API vs subscription covers when each makes sense.

What do tools, images and voice cost on the OpenAI API?

Web search costs $10.00 per 1,000 calls, a code container session costs from $0.03, and file search costs $2.50 per 1,000 calls. Tokens used by built in tools bill at the chosen model's rates as well.

Item, read 22 November 2026Price
Web search$10.00 per 1,000 calls, plus search content tokens at model rates
Web search preview, other models$25.00 per 1,000 calls, search content tokens free
Containers, 20 minute session$0.03 for 1 GB, $0.12 for 4 GB, $0.48 for 16 GB, $1.92 for 64 GB
File search storage$0.10 per GB a day, 1 GB free
File search tool call$2.50 per 1,000 calls
gpt-image-2.6-sunburst, Standard$8.00 image input, $30.00 image output
gpt-image-2.6-sunburst, Batch$4.00 image input, $15.00 image output
gpt-realtime-2.1, text$4.00 in, $24.00 out
gpt-realtime-2.1-mini, text$0.60 in, $2.40 out
gpt-live-1.1 voice$0.05 a minute active, $0.02 standby
gpt-live-transcribe$0.017 a minute

Container sessions are billed by the minute with a 5 minute minimum. OpenAI also says "Fine-tuning is closed to new users," so existing fine tuned GPT-4.1 models are the only ones you can still train.

What is OpenAI model pricing for ChatGPT and specialized models?

The specialized rows run from $5.00 to $20.00 per million input tokens, and none shows Batch or Flex. ChatGPT's own model, chat-latest, lists at $5.00 in and $30.00 out.

Model, Standard onlyInputCached inputOutput
chat-latest (ChatGPT)$5.00$0.50$30.00
GPT-6.2 Astra Law$8.00$0.40$40.00
GPT-6 Astra Law$10.00$1.00$50.00
GPT-6 Rosalind, life sciences access$10.00$1.00$50.00
GPT-6 Cyber, Daybreak access$20.00$2.00$100.00

Rosalind and Cyber are trusted access products, so most developers can't call them. The Astra Law and Rosalind promotional prices run at least through 31 December 2026. GPT-5.6 Sol has its own promotional price until 30 November 2026.

What are the best alternatives to the OpenAI API?

Anthropic and Google sell the closest alternatives, and at the top end both undercut GPT-6.2 Astra on list price. We read their pages on 22 November 2026. Claude Opus 5.5 lists $4 in and $20 out, half of Astra's $8 and $40. For a head to head, ChatGPT API vs Claude API pricing goes model by model.

Provider and modelInputOutputWhat to know
OpenAI GPT-6.2 Sol$2.00$10.00Batch $1.00 and $5.00
OpenAI GPT-6.2 Astra$8.00$40.00Batch $4.00 and $20.00
Anthropic Claude Sonnet 5.5$2$10Batch $1 and $5
Anthropic Claude Opus 5.5$4$20Batch $2 and $10
Anthropic Claude Haiku 5.5$0.10 up to 100,000 token prompts$0.50Batch $0.05 and $0.25 up to that size
Google Gemini 4 Argon$2.00$10.00Rises to $4.00 and $20.00 on 1 December 2026
Google Gemini 3.8 Flash$0.75$3.75Rises to $1.50 and $7.50 on 1 January 2027
Whizi ProNo APINo API$29.99 a month for chat, no tokens to count

Details live in our Claude API pricing and Gemini API pricing pages. Amazon Bedrock and Microsoft Azure also sell OpenAI models, and OpenAI says those are billed through the cloud platform, not through OpenAI.

Whizi can't replace any of these for an app, because it has no API and no API keys. It replaces the chat side, which is a different job. The next section says how.

Do you need the API, or one app for chat?

If you build software that other people use, you need the API. If you only use GPT, Claude or Gemini to write, check or analyze, one app costs less than a pile of keys, billing accounts and rate limit tiers.

Whizi Pro puts Claude, GPT and Gemini in one app, and its first week is $1.99. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card. Excel and Word files made in chat have no plan gate, and Pro and Powerhouse can compare two models on one prompt side by side. Whizi loses to the API on one point: your own product can't call it.

What mistakes make an OpenAI API estimate wrong?

Four mistakes cause most surprise bills, and each has a quick fix.

  1. Pricing only the input. The fix: output costs five times input on Sol and Astra, so a chatty answer dominates the bill.
  2. Copying a price from a blog. The fix: model names and rates changed several times this year, and the #2 Google result is a July 2025 post that lists GPT-4o, GPT-4 Turbo and GPT-3.5 Turbo, none of which appears on OpenAI's page today. Read the pricing page itself.
  3. Planning on Flex. The fix: the page says no new models will get it and retirement is coming, so use Batch for work that can wait.
  4. Forgetting the date. The fix: Ultrafast moves from 3 to 6 times Standard on 1 March 2027, and promotional prices end on 30 November and 31 December 2026.

Where does this pricing page fall short?

OpenAI's page is a list price, not your invoice. Our worked costs leave out tax, retries, cache write charges, the Regional processing uplift and any tokens a model spends reasoning. The page also has a short and long context setting. We read the short one, so check the long rates before you send huge prompts.

Every number that you see here came from the vendor pages that are named in the sources. Model names and prices have moved often this year, so treat all of it as a snapshot of 22 November 2026.

What can you ask an AI to check your OpenAI API cost?

Ask an assistant to do all of the multiplication for you, then check one line of it by hand, because that is where most of the errors hide. Paste any of these into Claude, ChatGPT, Gemini or Whizi, and fill the brackets.

  1. Estimate a bill: "My app sends [number] requests a month, each with [number] input tokens and [number] output tokens. Using the rates below, give me the monthly cost on Standard and on Batch. Show the arithmetic. Rates: [paste the rates]."
  2. Find the cheapest fit: "Here are three models with their input and output prices: [paste]. My job is [describe it]. Which is the cheapest that can do it, and what should I test to be sure?"
  3. Check caching: "My system prompt is [number] tokens and starts every request. With input at [price] and cached input at [price], how much do I save over [number] requests?"
  4. Audit a quote: "A blog says the model costs [price]. I paste the vendor page below. List every number in the blog that differs from the page: [paste]."
  5. Plan a launch: "I expect [number] users, each sending [number] messages a day. Give me a low, expected and high monthly cost, and a budget alert level for each."

Check the arithmetic against the rate card above before you trust it, since that is the part that most people skip. Models are good at the formula and sometimes slip on a decimal place.

Workflow checklist
  • Count tokens on one real request before you estimate anything
  • Test GPT-6.2 Sol before you pay for Astra
  • Send work that can wait through Batch at half the Standard rate
  • Cache any system prompt that starts every request
  • Skip Flex for new plans, since OpenAI says no new models will get it
  • Set a budget alert before launch
Common questions

Frequently asked questions

How much does the OpenAI API cost per token?

GPT-6.2 Sol costs $0.000002 per input token and $0.00001 per output token, as of 22 November 2026. That is $2.00 and $10.00 per million. GPT-6.2 Astra is $8.00 and $40.00. To see a real count for your own text, use the token counter, which shows token counts and an estimated API price.

What is the cheapest OpenAI API model?

GPT-6 Luna is the cheapest listed, at $0.05 in and $0.25 out per million tokens, but only under Flex, which is winding down. Among Standard rows the cheapest are GPT-6.2 Sol and GPT-6.1 Sol at $2.00 in and $10.00 out. Batch halves either one.

What is the difference between Batch and Flex?

Both price GPT-6.1 Sol at $1.00 in and $5.00 out, half of Standard, but Batch is the lasting option. OpenAI says Flex remains on its listed models while the beta winds down and will add no new models. Batch is listed on all four Standard models.

Is the API cheaper than a ChatGPT subscription?

It depends on volume. A light user can spend a few dollars on the API, while a developer sending 100,000 requests pays $900 on GPT-6.2 Sol. ChatGPT API vs subscription works through where the lines cross.