OpenAI Cost Calculator: AI Agent Cost per Month

This free OpenAI cost calculator shows your likely monthly bill. Pay per token for GPT models. No monthly fee. Your real cost depends on your own usage, so enter your numbers above. We use the official OpenAI pricing page, last checked on November 3, 2026.

What is the OpenAI API?

The OpenAI API lets your own software use the same GPT models that power ChatGPT. Instead of paying a flat monthly fee like a ChatGPT plan, you pay for exactly what your agent uses, measured in tokens. A token is a small piece of text, about three quarters of a word. Your agent pays for the tokens it sends to the model (input) and the tokens the model writes back (output).

Most AI agents, including ones built in n8n, Make, Vapi and many CRM tools, run on a model API like this one under the hood. So even if you never write code, the OpenAI price list often decides a big part of your agent's bill. This page shows what those prices mean in real monthly dollars.

How much does OpenAI API cost?

There is no monthly plan for the OpenAI API. It is pay as you go: you add credit or a card, and you are billed for the tokens you use. The price depends on which model you choose. Here are the main text models, per million tokens, with what one support chat costs on each. A support chat in our calculator is 3 model calls with 3,000 input and 400 output tokens each, and 50% of the input cached.

GPT-6 Luna is the budget model at $0.10 in and $0.50 out. GPT-6.1 Terra is the mid-price workhorse at $1.00 in and $5.00 out. GPT-6.1 Sol costs twice as much as Terra, and GPT-6 Astra is the top model at $10.00 in and $50.00 out. The GPT-5.6 Sol price shown is a promotional price that OpenAI says runs at least through November 21, 2026.

The jump between models is large. One support chat costs $0.0011 on GPT-6 Luna and $0.1095 on GPT-6 Astra, about 100 times more. Picking the right model matters more than anything else on your OpenAI bill.

Other ways OpenAI charges

• Flex and Batch: cheaper prices if you can wait for the answer. Good for reports and data jobs, not for live chat.
• Fast mode: faster answers at a higher price. OpenAI renamed "priority processing" to Fast mode in July 2026.
• Regional processing: if your data must stay in a certain region, newer models cost 10% more.
• Voice: GPT-Live voice sessions are billed at $0.05 per minute, with model and tool use charged separately.

What is included, and what is not?

The OpenAI price covers the AI model only. That is the "brain" of your agent. Here is what you get and what you still need to pay for:

Included in the token price

• Every model call your agent makes, billed per token.
• Function calling (tool use), where the model tells your code which action to take. You just pay for the tokens.
• The Responses API, Chat Completions API and Batch API. OpenAI does not charge extra for which API you use.

Not included

• Hosting for your agent code
• Phone numbers and voice minutes
• Web search calls ($10 per 1,000 calls) and other tools
• Code Interpreter and hosted shell containers, file search storage ($0.10 per GB per day after 1 GB free) and file search tool calls ($2.50 per 1,000).
• A no-code platform or framework if you do not write your own code.

What does OpenAI cost at 3 business sizes?

Here is what a customer support agent costs on the OpenAI API at three sizes, using GPT-6.1 Terra, 22 working days a month and no safety buffer. Add 20% to these for a safe budget.

The cost grows in a straight line with volume, because there is no plan fee. That makes the API easy to predict. At the large size, moving the easy chats to GPT-6 Luna matters a lot: running all 2,000 chats a day on GPT-6 Luna would cost $48.18 instead of $472.

Heavier agents cost more per task. A research agent doing 10 reports a day on GPT-6.1 Sol costs about $35.06 a month. A coding agent running 10 jobs a day on GPT-5.3 Codex costs about $135, because coding agents read a lot of code at every step.

What hidden costs should you watch for?

• Cache writes are not free. OpenAI's page lists a price to write text into the cache, as well as the cheaper price to read it. For GPT-6.1 Terra, a 30-minute cache write is $1.25 per million tokens, which is a bit more than the normal input price. Caching pays off only when the same text is reused many times. Our calculator uses the read price, so busy agents are estimated well and quiet agents may cost a little more.
• Long chat history. Every step resends the whole conversation. A chat that goes on for 20 messages can cost several times more than our average.
• Retries and loops. If a tool fails and the agent tries again, you pay again. A bug that makes an agent loop can burn through credit overnight.
• Built-in tools. Web search is $10 per 1,000 calls, plus the search text the model reads. A research agent that searches 10 times per report adds $0.10 per report before any tokens.
• Reasoning tokens. Some models "think" before they answer. Thinking tokens are billed as output, so a hard question can cost more than the visible answer suggests.

How can you lower your OpenAI bill?

1. Route by difficulty. Send simple messages to GPT-6 Luna and only the hard ones to GPT-6.1 Terra or GPT-6.1 Sol.
2. Put fixed text first. Keep your instructions and examples at the start of every prompt, in the same words, so they hit the cache.
3. Use Batch or Flex for anything that can wait. Nightly reports and data clean-up do not need instant answers.
4. Summarize old messages. Replace long chat history with a short summary after a few turns.
5. Cap output length. Set a maximum number of output tokens and ask for short answers.
6. Set a monthly budget. Use the usage limits in your OpenAI account so a bug cannot run up a surprise bill.

What are the main OpenAI alternatives?

Here is how the OpenAI API compares to other model APIs for the same support chat. All three alternatives are pay per token with no monthly fee.

• Anthropic Claude: Claude Sonnet 5.5 has the same list price as GPT-6.1 Sol ($2.00 in, $10.00 out), so a support chat costs $0.0215 on both. Claude Haiku 5.5 matches GPT-6 Luna at the low end. Many teams pick between the two on quality for their own task, not price.
• DeepSeek: DeepSeek Flash (V4.1) costs $0.0028 per support chat at peak hours, about 4 times cheaper than GPT-6.1 Terra, and half that again off-peak. Check whether its data and hosting terms fit your business before you switch.
• Google Gemini: also pay per token. We could not confirm Gemini prices on the official page at our last check, so it shows as "Price not confirmed" in the calculator for now.

Not ready to write code? You can still use OpenAI models through a no-code tool. See our n8n cost calculator to add an n8n plan on top of OpenAI tokens, or the Vapi cost calculator for phone agents.

Who is OpenAI best for?

What are the key OpenAI features?

• Many model sizes: From the low-cost GPT-6 Luna to the top GPT-6 Astra, so you can match cost to task difficulty.
• Responses API with tools: Built-in web search, file search and code tools the model can call during a task.
• Function calling: The model tells your code which action to take, like 'book_appointment', with the right details.
• Prompt caching: Repeated text at the start of a prompt is billed at a much lower cached rate.
• Batch and Flex: Cheaper prices for work that does not need an instant answer.
• Agents SDK: OpenAI's open-source library for building multi-step agents and handoffs between them.

How do you set up OpenAI?

The steps below show the general setup. Button names change often, so follow the idea of each step rather than exact labels.

1. Create an API account. Sign up on the OpenAI developer platform. This is separate from a ChatGPT subscription. Add a payment method and some credit.
2. Set a monthly budget. Before anything else, set usage limits and email alerts so a bug cannot run up a large bill.
3. Create an API key. Make a key for this project only and store it safely, for example in your server's environment settings. Never put it in website code.
4. Write your agent's instructions. Describe the job, the tone, what the agent must never do and when to hand off to a person. Keep the fixed instructions at the start so they get cached.
5. Define the tools. List the actions the agent can take, like looking up an order or booking a time, as functions your code will run.
6. Pick a model per step. Start with GPT-6.1 Terra for most steps and test GPT-6 Luna for simple ones like sorting messages.
7. Test with real examples. Run 20 to 50 real past conversations through the agent and check the answers before going live.
8. Launch and watch usage. Check the usage dashboard daily for the first week and compare it with this calculator's estimate.

Tips from setup

• Log the tokens used per task so you can spot expensive conversations.
• Cap the maximum output length for every call.

How do you calculate your OpenAI cost?

1. What will GPT do for you?. Pick the job closest to yours. We fill in typical usage for that job, so you do not need to know anything about tokens.
2. How many times a day?. Use a normal working day. A rough guess is fine, and we count 22 working days a month.
3. Which GPT model?. Lower-cost models suit simple work. Move up only if the answers are not good enough.
4. Read your answer. You see one monthly figure, the cost of each OpenAI task and what makes up the total.

Want to compare more tools at once? Use the main AI agent cost calculator.

How should you budget for OpenAI?

With a model API, you pay for what the model reads and writes, so your bill comes from your own design choices. The length of your instructions, how much chat history you resend, and how many steps each task takes matter as much as the price per token. Doubling the steps roughly doubles the cost.

To budget for OpenAI, run 20 to 50 real tasks through the model and record the tokens each one uses. Multiply the average by your expected monthly volume, then add 20% for retries and long requests. Set a monthly spending limit in your account so a bug cannot run up a large bill. Remember the API price does not cover hosting, your own development time or any paid tools your agent calls.

OpenAI pricing FAQ

Get your OpenAI cost estimate

Now put in your own numbers. The OpenAI cost calculator at the top of this page gives you a figure in about a minute. Then run a small test with real work and compare your real usage with the estimate. Still deciding? Compare other AI tools.

Source and last checked date

All OpenAI prices on this page come from the official OpenAI API pricing page, last checked on November 3, 2026. Standard tier, short context prices. Regional data residency endpoints cost 10% more for models released on or after March 5, 2026. GPT-5.6 Sol is on promotional pricing at least through November 21, 2026.

More cost calculators

Other ai model apis: Claude, DeepSeek, Gemini, Grok, Mistral.

All tools: n8n, Vapi, GoHighLevel, Retell AI, Bland AI, ElevenLabs, Synthflow, Make, Zapier Agents, Lindy, Botpress, Chatbase, Voiceflow, Relevance AI. Or go back to the AI agent cost calculator.