Cost per 1,000 Tokens vs per Million
AI pricing guide · updated 2026-08-07
Older pricing pages quote AI models per thousand tokens; current ones almost all quote per million. The arithmetic between them is trivial, which is precisely why the mistake is so easy to make: a figure that looks plausible can be out by three orders of magnitude.
The conversion
One million tokens is a thousand thousands, so a per-million rate is the per-thousand rate multiplied by 1,000. A model at $0.002 per 1,000 input tokens is $2.00 per million. Going the other way, divide by 1,000: $15.00 per million is $0.015 per thousand.
That is the entire relationship. The risk is not the sum; it is reading a number in one unit and assuming the other.
Why the industry moved to per-million
Prices fell far enough that per-thousand quotes became unreadable. The cheapest models we track blend to about $0.02 per million tokens, which is $0.00002 per thousand. Quoting four leading zeros helps nobody, and comparing two such figures at a glance is close to impossible.
Per-million also matches the scale at which people actually work. A million tokens is roughly 750,000 words, so monthly volumes land in units of millions rather than thousands. Providers standardised on the unit that produces numbers a reader can hold in their head.
Where the error shows up
The usual failure is a spreadsheet built from a documentation page in one unit and a competitor comparison in the other, producing a model that appears a thousand times cheaper than its rival. The figure is so far out that it tends to be caught, though not always before a decision rests on it.
The subtler version is an old blog post or cached comparison quoting per-thousand rates from an earlier era. Those numbers are usually stale in both unit and value, since prices have moved since.
Guard against both by reading the unit off the provider's own current pricing page, and by sanity-checking any figure that implies a monthly bill wildly different from what a similar workload costs elsewhere.
Frequently asked questions
How do I convert AI token prices between units?
Multiply a per-1,000-token price by 1,000 to get the per-million price, or divide a per-million price by 1,000 to get per-thousand. So $0.003 per 1,000 tokens is $3.00 per million, and $10.00 per million is $0.01 per thousand.
Why do providers quote per million tokens?
Because prices fell far enough that per-thousand figures became unreadable. The cheapest models we track work out at around $0.00002 per thousand tokens, and comparing numbers with that many leading zeros is impractical. A million tokens is also close to the volume real applications consume, at roughly 750,000 words.
How many words is a million tokens?
About 750,000 words of ordinary English prose, since a token averages roughly four characters. Code, other languages and heavy punctuation all tokenise into more tokens per word, so treat that as a planning figure rather than an exact conversion.
See it in the data
Related guides
Terms used in this guide
Published by Tokenando. Last updated 2026-08-07. Figures in this guide are computed from our own pricing index and dated where they can move; see the methodology and corrections policy.