The adoption score (0–100)
Tools are ordered by an adoption score from 0 to 100. It is an editorial judgement of how widely a tool is used in practice, set by hand for each tool. It is not computed from a formula.
When setting it, we look at:
- Search demand: how often developers search for the tool by name.
- Downloads and installs: package registries such as npm and PyPI, the VS Code Marketplace and Docker Hub, where the tool is distributed that way.
- Open-source activity: GitHub stars and their growth, contributors and dependent repositories, for open-source tools.
These signals are weighed by judgement, not by fixed weights, and the score does not yet record which figures were used for each tool. Treat a gap of a few points as a tie. Scores change when a re-check finds that a tool's use has clearly grown or shrunk, and every change shows up in the tool's history.
How prices are checked
The entry price is the cheapest paid plan available to one developer, such as $10/mo or $20/mo. Open-source tools you can run yourself are listed as Free, and pay-as-you-go services as Usage. Tools sold only through a sales contract are listed as Enterprise only.
Prices are checked in rounds rather than on a fixed schedule. In each round an editor opens the vendor's own pricing page and compares it with the index. So far the index has been checked on Oct 13, Oct 15 and Oct 29, 2026.
- Confirmed: the price was read on the vendor's own pricing page. Each tool page shows the date of that check.
- Not yet confirmed: the price comes from other public sources and has not been read on the vendor's page yet.
- Under review: the official price could not be read, for example after an acquisition or a redirect, so the price reads Under review until it can be.
Acquisitions, rebrands and shutdowns found during a check are noted on the tool's page with the date they were recorded.
Models and benchmarks
Model API prices are read from each provider's own pricing page and stored with the date they were read. The blended price is (3 × input + output) ÷ 4, per million tokens.
Benchmark scores come only from independent evaluators: Vals AI, Artificial Analysis, Arena and Epoch AI. Vendor self-reports are not used. Each score keeps its evaluator's own unit and the exact configuration measured, such as a reasoning setting or a dated snapshot. Scores are copied from each evaluator's published leaderboard on the date shown, not run by us.
The long-run charts on AI in charts use Epoch AI's Notable AI Models dataset (CC BY 4.0) as published, including its estimates of training compute. Trend lines there are least-squares fits that we compute from that data.
A model appears in a ranking only once its price has been confirmed, so each ranking says how many of the evaluator's models it covers, and names the evaluator's leader when that is a model we have not priced.
History
Every update to the data is saved as a dated snapshot of each tool's rank, adoption score, entry price and license, and each model's price. Rank movement and price changes on the site are computed by comparing snapshots, and the snapshots are published as open data at /data/history.
Editorial independence
- Rank cannot be bought. No vendor can pay to change its adoption score or its position.
- Links are followed. Links to listed tools are ordinary, followed links.
- Paid placement, if it ever exists, will be labelled and marked
rel="sponsored". A paid review would buy review turnaround time only, never inclusion or rank.
Badges
Every listed tool has an SVG badge at https://theindexofai.com/badge/<slug>.svg showing its current category rank, for use in a README, a landing page or documentation. Badges are regenerated whenever the index is rebuilt, so they follow the current ranking.
Corrections
Anyone can suggest a correction to a price, a license or a description. A correction changes the data only after it has been checked against the vendor's own page.
Send a correction, or write to editorial@theindexofai.com.