How we research and test

Every number on toolist.ai links to the page it came from and carries the date we checked it. Each page says at the top whether it is researched from public sources or hands-on tested by us, and when it was last updated.

AI models

  1. Prices come from the provider. We read each provider's own pricing page (OpenAI, Anthropic, Google, Meta) and cross-check it against OpenRouter's public model list. We re-check model prices every week.
  2. Limits come from the model's own documentation: context window, maximum output, knowledge cutoff and release date.
  3. Benchmarks come from the provider or Artificial Analysis only, with the settings they were measured at. We do not publish estimated scores.
  4. Monthly costs are our own calculation. We price the same three workloads (a support chatbot, document Q&A and a coding agent) at standard rates, and the calculator on each comparison lets you enter your own token counts.
  5. Same-prompt tests only when we run them. When we send the same prompts to two models, the page shows the outputs with the date and settings. Until then, a comparison has no test section.

AI tools for video, voice, images, writing and coding

  1. Prices come from the vendor. We read every plan on the vendor's pricing page, billed monthly and yearly, and record the day we checked. Every week our price watcher reads each tool's pricing page again; when a price has moved, an editor re-checks it on the page and the change goes into the tool's price history.
  2. Price per unit is our calculation. We divide a plan's monthly price by the minutes, images, words or credits it includes and label the result as calculated, not measured. The cost calculator uses the same published allowances.
  3. Hands-on findings come from named testers. We collect published reviews whose authors used the tool, such as TechRadar, Zapier and PCMag, and summarize what they found in our own words, with a link and the date. We skip listicles that never used the tool and sponsored posts.
  4. User ratings come with their counts, from review sites such as Trustpilot and Capterra.
  5. We say what we did not do. We have not run our own tests of these tools. A page says "tested by toolist.ai" only if we ran the test, and then it shows the date and the output.
  6. New tools are listed before they are researched. A listed tool says "not researched yet", shows no prices, and stays out of search engines until an editor has researched it and a second editor has checked every number.

How AI helps

Our editors use AI tools to research and draft pages. Before a page goes live, a second reviewer checks every price, limit, benchmark and date against its source, and we record that check. AI never supplies a number, a score or a test result: if we could not find a number at its source, the page leaves it out.

Corrections

If a number is wrong or out of date, email hello@toolist.ai. We fix it, re-check the page and update its date.