How we rate AI tools

We compare each tool to its direct rivals. Every tool is scored against others that do the same main job, such as coding, voice or video. If a tool also fits other tasks, we list them, but its score comes from its main job.

Four criteria, each scored 1 to 5:

• Quality (40%): how good the output actually is. We use blind-vote leaderboards such as LMArena and Artificial Analysis, but only when the product really runs the model being ranked. We also use hands-on reviews from outlets like PCMag, agreement across independent reviews, app store ratings, and the health of open-source projects. If a leaderboard lists a product by name, we count that result, good or bad. A tool that isn't on a leaderboard is never marked down for it.
• Features (25%): we build a checklist for each category from what matters for that job. For coding, that includes agent mode, multi-file edits, terminal access, PR review, model choice, IDE support and privacy options. We tick an item only if the tool's official docs show it.
• Value (20%): price and free-plan limits compared with the typical rival in the same category.
• Ease (15%): how quickly a typical person gets going: skill needed, sign-up friction, free plan, and what testers say.

Why these weights: output quality is what you actually get, so it counts most. Features decide whether a tool can do your job at all. Price matters but shouldn't win over a tool that does the job better. Ease matters most to beginners, and our quiz already filters for skill level.

Quality floor: if a hands-on tested review scores a tool 2 out of 5 or lower, it can't be #1 in its category, however cheap or feature-packed it is. It's placed just below the best tool that clears that bar.

Ties: tools within 0.1 of each other are shown as tied.

What the numbers mean: 5 is rare and best in class. 4 is clearly above average. 3 is average for the category. 2 is weak. 1 is poor. We never stretch or curve the results to make them look spread out. The numbers come straight from the evidence.

Evidence, not opinion. We keep the evidence and sources for each score on file and review them quarterly. Those records include leaderboards, official pricing and docs, tested reviews, app store pages and GitHub. We read forums for leads but never treat them as fact.

Money never buys a score. Some links earn us a commission. That has no effect on ratings, rankings or which tools we cover.

Unrated tools. If we can't find at least one tested review or measurable signal plus verified pricing, the tool stays listed without stars until we can.

Updates. We check prices every week and review every rating at least once a quarter. We re-score sooner after a major release, a price change, or signs a project has stopped being maintained. Each rating shows its last review date.