How I score AI tools

What I test

• Real tasks, not vendor demos: the tool used in actual workflows.
• Pricing verified against official sources and refreshed regularly.
• Capability boundaries: where the tool shines and where it breaks.
• Live signals: model pricing and GitHub stars, tracked over time.
• Trade-offs: every tool has them, and the review names them.

The rating

Scores are out of 5 and reflect overall value for the tool's intended user, not a feature checkbox count. A high price is fine if the tool earns it; a free tool still loses points if it wastes your time. The verdict spells out who should use it and who should skip it.

What I don't do

No tool pays for a higher score or a better spot in a ranking. Affiliate links are disclosed and kept out of the rubric. I don't cherry-pick the one benchmark that flatters a tool, and I don't tuck the weaknesses below the fold. When a popular tool is overrated, the review says so.

Live data

Pricing and popularity figures are pulled from public APIs and refreshed daily, so reviews stay current instead of decaying. Where a number appears, it reflects the most recent sync.

When a score changes

Reviews aren't frozen. A tool can climb when it ships a genuinely better release, or slip when a price hike or a quality regression changes the math. The live pricing and popularity numbers update on their own; the written verdict gets revisited when a tool has changed enough to matter. Each review shows when it was last looked at, so you can judge how current the call is.

Versions move fast

AI tools ship breaking changes on a weekly cadence, so a review pins the version and the date it was tested. A verdict on last quarter's release is labelled as such, not passed off as current. When a new version lands that changes the experience, the review is re-run against it rather than patched from the changelog.