About this index

Price per 1M tracks what AI models actually cost. The index currently covers 1,657 endpoints from 32 providers, normalising everything to a single unit — US dollars per one million tokens — so two models on different pricing pages can be compared without a spreadsheet. From that dataset we publish 45 curated model pages, 10 use-case rankings, 90 head-to-head comparisons and a weekly price report.

Who runs this

Price per 1M is built and maintained by Mohammad Ghonchegi, an independent developer who needed an answer to a recurring question — “what will this model actually cost me per month?” — and could not find one that stayed current. It is not backed by a provider, an agency or a venture round. The same person who writes the code answers the inbox, so corrections go straight to the source.

Why it exists

Model pricing changes constantly: cuts, new tiers, cache discounts, retirements. Provider pricing pages are authoritative but not comparable, and most comparison articles go stale within weeks. This site is built as a pipeline first and a website second — data is refreshed automatically, and every page is generated from the same dataset, so nothing drifts out of sync and nothing is hand-copied from someone else's table.

How the data is collected and checked

1. Collection. Rates are read from provider pricing pages and from two community-maintained price registries, then normalised into one schema (input, output, cached input, context, max output, capability flags).
2. Second source. Where a second registry or the provider's own page exists, the figure is cross-checked; a disagreement above a couple of percent is shown on the page with both numbers rather than averaged away.
3. Change tracking. Every refresh is diffed against the previous snapshot, which produces the changelog and the weekly report. Nothing is silently overwritten.
4. Publication. The dataset is published as JSON (API), as an open dataset mirror, and rendered into the pages you are reading. The full recipe is on the methodology page.

Editorial policy

• Only published, attributable prices. If a provider does not publish a rate, we show a dash rather than an estimate.
• No provider pays for placement, ranking or a favourable comparison, and no provider sees a page before it is live.
• Rankings are computed from the dataset with the formulas published on the methodology page, so anyone can reproduce them.
• No scraped third-party copy: page text is written for this site, and the numbers are produced by the pipeline.
• Estimates are labelled as estimates, and exclusions (batch, committed-use, regional, free tiers) are stated on the page.

Editorial articles and review

The data pages here are generated; the guides are not. Those articles are written by hand around the numbers and reviewed against the dataset before publishing, and every figure inside them is interpolated from the same source that renders the tables — so an article cannot drift away from the data it quotes. 10 articles are published so far, covering cost planning, prompt caching, cross-platform price spreads and the pricing terms that rate tables leave out. Where an article makes a judgement rather than a calculation, it is labelled as a judgement.

Why some pages are deliberately not in search

One model is often resold by a dozen platforms, so the raw dataset would produce a dozen nearly identical pages per model. Instead you get one canonical page per model — which lists every platform's price side by side — and the duplicate listings stay reachable from that page without being advertised to search engines. It keeps the index useful instead of voluminous.

Accuracy and corrections

Data is refreshed on an automated schedule; the last successful refresh was 2026-10-28 12:57 UTC. If you spot a price that is wrong or out of date, tell us via the contact page and we will correct the pipeline — corrections are usually live within one refresh cycle. Providers who want a rate attributed differently, or a retired model marked as such, can request it the same way.

How the site is funded

By display advertising, and potentially by sponsorship of a specific data page, which would be labelled as such. Ads are served programmatically, they load only after you agree to advertising cookies, and they have no influence on the data or on the order of any ranking. See the terms for the formal statement.

What you can do here

• Compare per-million-token input, output and cached-input prices across providers.
• Estimate a real monthly bill with the cost calculator.
• See what changed and when in the pricing changelog.
• Read the same dataset programmatically from the JSON API, llms.txt or llms-full.txt.

Price per 1M is an independent publication and is not affiliated with, endorsed by, or sponsored by any of the providers listed. Model and product names belong to their owners and are used only to identify the products compared.