TOKENHOURS converts every major model's token price into an honest hourly rate, in your currency, and puts it next to what a human hour costs you. Then it shows you the part nobody budgets for: MCP servers, API calls, and CLI agents.
New here? Start at the top and read down. Or press Ask The Guide and just talk to it.
You do not need a tech background to use this site. You need five ideas. Each one takes about thirty seconds. Read them once and the rest of the page will make sense.
AI does not read words. It reads small chunks called tokens. One token is about ¾ of a word. So 1,000 words is roughly 1,300 tokens. A single page of text is about 500 tokens. This whole page you are reading is around 6,000.
What you send in is called input. What the AI sends back is output. Output costs about five times more, because writing is harder work than reading. Shorter answers save you more money than shorter questions.
Companies list prices like "$3 per million tokens." That sounds tiny, and that is the trap. A busy AI helper can burn a million tokens in an hour. The number only means something once you know how fast you are using it.
AI has no memory between messages. Every time you reply, the whole conversation gets sent again. Message ten costs far more than message one, because it carries messages one through nine with it. This surprises almost everyone.
Caching means the AI remembers the repeated part of your request and charges about a tenth for it. Batching means you accept a slower answer and pay half. Both are on the calculator below.
Still fuzzy on any of it?
Type or talk. It knows this whole page.
Every AI model you can rent, side by side. Input is what you send. Output is what it writes back. Cache hit is the discount for repeated text. Blended is one fair number for comparing models quickly.
| Model | Input | Output | Blended |
|---|
"Memory" is how much text the model can hold at once, in tokens. Rows marked LIVE were pulled from a public price feed when you opened this page. Rows marked CHECKED were verified by hand from each company's own pricing page on 29 July 2026. Prices change often. Confirm with the provider before you commit real money.
Set what an hour of a person's time costs you. Pick your currency. Choose the kind of work. The page works out what the AI costs for that same hour, and how the two compare.
Their wage, or what you bill a client for the hour
Every price on this page converts from US dollars. AI companies still bill in dollars, and your bank adds its own small fee on top.
This is the most important setting on the page
Repeated text gets the cheap cache price
Careful: saving text to the cache the first time costs a little more than normal. If very little repeats, caching can cost you instead of saving you.
Click any row to price it above
Describe what you want to build in your own words. The Guide reads it, works out which models it needs, which connectors it needs, how many rounds of rework it will take, and how many developer hours that adds up to.
You get one number: the real cost to build it. Tokens, connectors, and human hours, all in real dollars.
This estimator is a tool to assist with cost planning. These figures do not take into consideration additional prompts, iterations, style preferences, or the other end goals of the builder. Every build is different, and your actual cost will differ from what you see here. Use the range as the answer, not the single number, and measure your own real usage as soon as you begin.
Plain English. The more detail, the better the estimate.
Your own time counts. Put what an hour of it is worth.
It cannot see how many extra prompts you will send, how many times you will change your mind, how particular you are about how things look, or what you are really trying to achieve in the end.
Those four things are usually the difference between the estimate and the invoice. That is why you get a range.
Nine projects people actually build. Each shows what the AI bill looks like to build it, and what it costs to keep running every month. Change your currency or your discount settings above and these all update.
Build cost is not project cost. These numbers are the AI bill to get a working thing. Web hosting, a domain, a database, and your own time all sit on top.
The monthly cost is the one that hurts. A $40 build that costs $900 a month to run is a worse business than a $4,000 build that runs on $60.
Triple your first guess. First builds always take more tries than planned. Every failed try still costs money. Budget three times what you think.
When you open the page, TOKENHOURS pulls a public price feed for live rates. Anything the feed does not cover falls back to prices we checked by hand on 29 July 2026 from each company's own page. Every row says which one it is, so you always know.
A free daily exchange-rate feed, covering around 160 currencies. If that call fails, a stored table loads instead and the badge at the top says so. Converted prices are for planning. AI companies bill in US dollars, and your bank's fee is not in this math.
An AI hour means one hour of steady work at the speed you picked. It is not one hour of the clock running. Speed is the honest variable, which is why the Kind Of Work setting matters more than the model you choose.
It will not quote your exact bill. It does not cover image, voice, or search pricing, and it does not know about volume discounts your company may have negotiated. Use it to plan, not to invoice.
The token amounts in Section 03 are estimates built from typical project runs, not audited measurements from a specific build. They are honest ballparks. Treat them as a starting range and measure your own once you begin.
Price the work. Price the way you build it. Then decide.
Open The Calculator