AI model prices,
compared honestly.
Per Million tracks live pricing across every major AI vendor and host, and normalizes it to one number you can actually compare: dollars per million tokens. No spreadsheets, no guessing which price page is current.
Every vendor prices tokens differently. Per Million doesn't.
Input and output tokens are billed separately, at different rates, on pages that change without notice. Comparing two models today usually means five browser tabs and a calculator.
Fragmented pricing pages
Every AI company publishes its own rates, in its own format, on its own schedule — some behind a console login.
Input ≠ output ≠ one price
A model's "price" is really two numbers plus a ratio assumption. Most comparisons quietly pick one and ignore the other.
Same model, different price
The exact same model is often resold through several hosts at different rates — and nothing tells you which one is cheaper today.
One pipeline, run every day
The catalog isn't hand-maintained — it's rebuilt automatically, so a rate change shows up here without waiting on an app update.
Collect
Pricing is pulled from every major vendor and host, once a day, from public catalog sources.
Normalize
Input and output rates are blended into one honest per-million-token figure, using a stated 3:1 ratio — never hidden math.
Cross-reference
Models sold through more than one host are linked together, so a cheaper copy of the same model is never invisible.
Publish
Your app fetches the refreshed catalog automatically. Nothing to sync, nothing to configure.
Built to answer one question fast: what will this actually cost?
Browse every model
Search, sort by price, context window or recency, and filter by vision, tool calling, reasoning, or free-only.
One blended price
Input and output rates combined into a single, clearly-explained per-million-token number — never a guess.
Also available on
See every host reselling the same model, with price, tier, and region side by side — so a cheaper copy never hides.
Compare side by side
Line up two or more models to see how price, context and capabilities actually stack up.
Cost calculator
Enter your own expected token volume and see what a model would really cost you, not just its headline rate.
Sign up where it's sold
Every vendor and host links straight to the real page to get access — never a redirect that's gone stale.
What "blended price" means
Vendors bill input and output tokens separately, so quoting "the price" of a model is already a small act of interpretation. Per Million is upfront about which one it makes: a typical 3-to-1 input-to-output ratio, stated in the open.
blended = (input × 3 + output) / 4Seventeen companies. One place to compare what they charge.
From the labs that train frontier models to the clouds and gateways that resell them — Vendors shows how many models each one offers, and what they start at.
No account. No tracking. Just the catalog.
Nothing to sign up for
Open the app and the catalog is there. No account, no email address, no login screen standing between you and a price.
Refreshed automatically
The catalog is rebuilt once a day from public sources. There's no stale local copy pretending to be current.
Your data stays on your device
Nothing about you is collected. Read exactly what that means on the privacy page.
Stop guessing what a model costs.
Per Million is available now on iOS and coming soon to Android. Get the app today.