ChatGPT Credits for Normal People: What Does AI Work Actually Cost?

If you have looked at OpenAI’s new ChatGPT credit rate card and thought, “Okay, but what does any of this actually COST me?” you are not alone.

Input tokens. Output tokens. Cached tokens. Credits per million tokens.

Perfectly logical if you build AI systems for a living.

Not terribly helpful if you run a sales department.

So let’s translate ChatGPT credits into something businesspeople actually understand:

Jobs.

How much does it cost to have AI perform a job?

That is becoming an increasingly important question because ChatGPT is not just a chatbot anymore. It can research, work in spreadsheets, build presentations, write code, navigate multi-step workflows and increasingly perform work on our behalf.

And that changes how we need to think about AI costs.

First: Not Everything You Do in ChatGPT Costs Credits

This is probably the most important thing to understand.

According to OpenAI’s current Business and Enterprise/Edu rate card, regular Instant ChatGPT messages are unlimited. More powerful GPT-5.6 Sol messages use approximately 10 credits each, while GPT-5.6 Sol Pro and GPT-6 Pro messages use approximately 50 credits each.

Then there are several features with relatively straightforward credit rates:

What you’re doing	Approximate usage
Instant Chat	Unlimited
GPT-5.6 Sol message	10 credits
GPT-5.6 Sol Pro message	50 credits
GPT-6 Pro message	50 credits
Agent Mode message	30 credits
Deep Research task	50 credits
Generate an image	5 credits
ChatGPT Voice	1.25 credits per minute

Those are the easy ones.

Then we get into agents, ChatGPT Work, Codex, Excel and PowerPoint.

That is where people start reaching for the Advil.

Why Agent Work Is Different

Imagine asking ChatGPT:

“What are three good restaurants near our hotel?”

That is basically a question.

Now imagine saying:

“Research restaurants near our hotel, compare ratings and menus, check which ones have availability Friday night, create a shortlist based on our preferences and put the recommendations into a document.”

That is a job.

The AI may need to search, read multiple pages, compare information, reason through your requirements, use several tools and then create something for you.

The amount of computing required will not be identical every time.

That is why OpenAI meters much of ChatGPT Work, Workspace Agent and Codex activity based on how much information the AI processes and produces.

OpenAI calls those units tokens.

For normal humans, I suggest thinking about them as:

Stuff the AI has to read + stuff the AI has to think through + stuff the AI has to create.

More stuff generally means more credits.

Here’s the Better Way to Think About Credits

Do not ask:

“How many tokens will this use?”

Ask:

“What does this business outcome cost me?”

OpenAI says a typical end-to-end Workspace Agent run using GPT-5.6 may consume approximately 5 to 25 credits.

A typical Codex task using GPT-5.6 Sol may consume approximately 5 to 30 credits.

Those ranges are much more useful for planning than trying to count tokens before the work has even started.

Let’s imagine a sales manager has an agent that does this every morning:

“Review yesterday’s sales activity, identify opportunities that have not been touched, flag deals that appear stalled and give me my five highest-priority follow-ups.”

Suppose a typical run consumes 15 credits.

Run it every business day:

15 credits × 22 days = 330 credits per month.

Now we have something we can evaluate.

If that agent saves the sales manager 20 minutes every morning, that is more than seven hours returned to that person every month.

Suddenly, 330 credits is not the interesting number.

Seven hours is.

A Few More Examples

Consider an executive who wants a weekly competitive intelligence report.

The agent searches for developments involving five competitors, reviews relevant sources, summarizes meaningful changes and produces an executive briefing.

If a run averages 20 credits:

20 credits × 4 weeks = 80 credits per month.

Or consider an operations manager using ChatGPT for Excel to analyze a large operational spreadsheet.

OpenAI says a typical ChatGPT for Excel or Sheets task may consume approximately 5 to 20 credits per message, although actual consumption depends on the amount of information being processed and generated.

Ten meaningful spreadsheet analysis jobs at an average of 12 credits would consume approximately:

12 credits × 10 jobs = 120 credits.

Maybe your marketing team creates presentations.

OpenAI estimates a typical ChatGPT for PowerPoint task at approximately 10 to 50 credits per message.

If creating and refining a monthly presentation takes four substantial interactions averaging 25 credits each, that would be approximately:

4 interactions × 25 credits = 100 credits.

Again, do not stop at the credits.

Ask what the human alternative was.

If somebody previously spent three hours assembling that presentation and AI reduces the job to 30 minutes of review and refinement, you are buying back 2.5 hours of skilled human capacity.

That is the number I care about.

Why Can the Same Job Cost Different Amounts?

AI work is not like buying photocopies.

One job can be dramatically larger than another.

Ask an agent to summarize a two-page document and it has relatively little information to process.

Ask the same agent to analyze 40 documents, search the web, compare the findings with a spreadsheet and produce a 20-page report, and you have handed it considerably more work.

OpenAI’s token-based system accounts separately for input, cached input and output.

Did your eyes glaze over again?

Mine too.

Here is what matters:

Reading costs something. Reusing previously processed information can cost less. Creating new material generally costs more.

Bigger and more complicated jobs usually consume more credits than smaller jobs.

That is enough information for most business users.

What Does a Credit Actually Cost?

This is where businesses need to look at their own billing arrangements.

OpenAI publishes how many credits different models and activities consume, but there is not one universal dollar price that applies to every Business, Enterprise and Edu agreement.

Business workspace owners can check their purchased-credit price under:

Workspace Settings > Billing

Enterprise and Edu pricing may depend on the organization’s contract or order form.

Once you know your effective price per 1,000 credits, the calculation is simple:

Monthly credits ÷ 1,000 × price per 1,000 credits = estimated monthly AI cost

If your effective cost were $10 per 1,000 credits and a workflow used 330 credits per month:

330 ÷ 1,000 × $10 = $3.30 per month

That $10 figure is simply an illustration. Use the actual rate shown in your workspace or agreement.

The Formula Businesses Should Actually Use

When I am helping companies evaluate AI workflows, I am much more interested in this:

Monthly AI Cost ÷ Monthly Hours Recovered = Cost Per Human Hour Recovered

Then compare that with:

Fully Loaded Human Cost Per Hour

Here is a hypothetical example.

An employee earning $80,000 annually has a fully loaded employment cost higher than salary alone.

Suppose their effective cost to the organization works out to $55 per hour.

An AI workflow saves that employee 10 hours every month.

That represents:

10 hours × $55 = $550 of capacity potentially recovered each month.

Now determine what the AI workflow costs to operate.

If the cost is $20, $50 or even $100 per month, you have a pretty interesting conversation.

And remember, time is not the only return.

An agent might also:

• Reduce errors

• Shorten response times

• Enforce a process more consistently

• Catch missed opportunities

• Improve customer experiences

• Let employees spend more time on higher-value work

Those benefits may be harder to measure, but they still matter.

“How Much Does AI Cost?” Is the Wrong Question

We have spent decades buying software primarily by the seat.

$30 per user.

$70 per user.

$150 per user.

AI is starting to look different.

We are increasingly paying not simply for access to software, but for work performed by software.

That is an important distinction.

If an employee uses ChatGPT to ask questions throughout the day, that is one usage pattern.

If an AI agent independently performs a 20-minute workflow that previously required an employee to complete it manually, that is something else entirely.

Businesses will increasingly need to evaluate AI using something that looks surprisingly similar to labour economics:

• What did the work cost?

• How long did it take?

• How often do we perform it?

• What would it have cost a human to perform?

• What is the value of the output?

• What human review is still required?

• Should a human have been doing this work in the first place?

Start With Workflows, Not Credits

This is why I keep coming back to workflow assessment.

Buying a giant bucket of AI credits and telling employees to “go use AI” is not much of a strategy.

Instead:

1. Identify repetitive, tedious, error-prone and time-consuming workflows.

2. Measure how much human time they consume today.

3. Determine which parts AI can reasonably perform.

4. Estimate the review time that will still be required.

5. Run the workflow enough times to establish average credit consumption.

6. Calculate the monthly AI operating cost.

7. Compare that cost with the capacity and other benefits recovered.

Now AI consumption becomes measurable.

You can even get surprisingly granular.

Invoice-routing agent

Runs: 800 per month
Average credits per run: 8
Monthly credit consumption: 6,400 credits
Human time before AI: 65 hours
Human review and exception handling: 12 hours
Capacity recovered: 53 hours

That is the beginning of a business case.

“1.7 million tokens” is not.

I Built a Calculator to Make This Easier

I realized that explaining the formula was only half the job.

So I built a ChatGPT Credit Calculator for Normal People.

You enter:

• The workflow you want AI to perform

• How often it runs

• Average credits consumed per run

• Your actual purchased-credit price

• Manual time required today

• Human review time after introducing AI

• Fully loaded human cost per hour

The calculator automatically estimates:

• Monthly credit usage

• Monthly AI operating cost

• Human hours recovered

• Capacity value

• Net monthly value

• Cost per human hour recovered

• Estimated ROI

• Annual net value

It also includes example workflows and a quick-reference sheet showing the current credit ranges for common ChatGPT tasks.

The goal is not to create a perfect financial forecast on day one.

The goal is to help you decide whether a workflow is worth testing and identify what you need to measure during the pilot.

One Last Thing: Do Not Memorize the Rate Card

These numbers will change.

OpenAI’s current rate card already includes promotional pricing, changing model availability and different rules for included versus purchased usage.

Bookmark the official rate card instead:

OpenAI ChatGPT Business and Enterprise/Edu Rate Card

Then build your business cases around:

Workflows, frequency, average consumption and value created.

Because the future of AI budgeting probably is not:

“How many ChatGPT licences do we have?”

It is:

“How much work are our humans delegating to AI, what does that work cost us, and is it worth it?”

That is a question normal people can answer.

Want the Calculator?

I am making the ChatGPT Credit Calculator for Normal People available as a downloadable Excel workbook.

I am not putting a public download link here because I would like to know who is using it and what kinds of workflows they are trying to evaluate.

Message me “CREDIT CALCULATOR” and I will send you a copy.

And if you are having trouble deciding which workflows to calculate in the first place, that is exactly where an AI Workflow Assessment can help.