How CheckAI compares AI tools

CheckAI does not run its own quality tests and never invents data. Everything you see comes from a source with a check date, and anything we could not confirm is labelled as such.

Where the data comes from

Every price, usage limit, model and feature in our catalog points to a source - usually the vendor's official pricing page or help center - together with the date we checked it. Product pages list these sources.

Prices are US list prices before tax unless the page says otherwise. Prices in other countries come from separate regional tables and are shown as reference prices.

How fresh the data is

Each kind of fact has a shelf life: prices 7 days, usage limits 7 days, features 30 days, exchange rates 7 days.

When a verified price is older than that, it is shown as "re-check due" until we check it again. Prices we could not confirm are marked "needs check". Unknown prices are never shown as zero.

Unknown is never "yes"

A feature is "yes" or "no" only when a source confirms it. Everything else is "not confirmed" - it never counts as included, never wins a comparison and never makes a plan "enough" for a task.

Comparing two plans

The comparison puts one plan of each tool side by side: price, price per included unit, usage limits, models, context window, features and external benchmarks.

Features are weighted by category: for coding tools an agent that edits your project matters more than voice mode; for image tools commercial terms and private generations matter most. Key features are listed first, and the weighted feature score counts only key features that are confirmed.

Task verdicts

For common tasks - everyday questions, writing, research with sources, documents, coding, images, video, voice and more - we check which plan has every capability the task requires, then the extra capabilities that help.

A plan with a confirmed required capability beats a plan where it is missing or not confirmed. If both are equal, the plan with more confirmed helpful capabilities wins; otherwise it is a tie. The verdict names the capabilities that make the difference.

Your workload

Pick a task and a volume - for example 20 minutes of video a month - and we convert it into the units each plan actually counts (seconds, credits, characters, runs), compare it with the included quota and hard limits and show "enough", "not enough" or "not confirmed".

The first-year cost includes known extra packs. External costs we cannot price, such as hosting or model API usage, are named and keep the total marked as "at least".

Price per unit

We divide the monthly price by the monthly quota included in the plan. Daily and one-time allowances are never multiplied into a monthly number.

Units of different vendors - for example credits - are never compared with each other. A lower price per unit is highlighted only for universal units such as characters or minutes, or between plans of the same service.

Models and external benchmarks

We list the models and context window each plan includes, as stated by the vendor. If vendor pages disagree, we show "sources disagree".

External benchmarks are a separate layer: Arena text score (https://arena.ai/leaderboard/text, updated 2026-10-02); Artificial Analysis Intelligence Index (https://artificialanalysis.ai/leaderboards/models, updated 2026-10-07). We show the exact leaderboard model, the source and its update date. When the vendor does not confirm which model version a plan uses, we say so. Benchmarks measure models, not your experience with an app - limits and features still matter.

Editorial scores

Scores from 0 to 10 on tool pages are CheckAI's editorial judgement of the tool overall, based on confirmed features, published limits, price and public benchmarks. They are not lab tests and do not describe a specific plan.

Editorial scores are shown separately in comparisons and do not affect task verdicts, the weighted feature score or the AI Finder.

Money and rankings

Some links on the site are affiliate links. Commissions never affect comparisons, verdicts, scores or recommendations.

Catalog version 2026-10-07. Compare AI tools side by side · AI Finder