How we rank AI models and tools

How the score is calculated

Each entry's score is the weighted sum of the sub-scores below (each 0–100). Models and tools use the same formula, but each is compared only with its own kind.

Performance

55%

Models: capability on public benchmarks and evaluations — we publish only our normalized sub-score, never raw benchmark numbers. Tools: output quality and feature depth for the task, not benchmarks alone.

Real-world reputation

35%

Community reviews, expert write-ups and adoption signals, gathered with AI research tools and checked against cited sources.

Recency

10%

Frequency of recent updates and new-feature cadence. A model whose own successor has shipped is capped at 75 for recency, and a retired model at 40.

Price — shown on every entry (free plan and paid plan prices) but not part of the score, so a product doesn't rank higher just for being cheaper.

Weights apply when an entry is re-measured in a research update. Entries not yet re-measured keep their previous score until their next update.

Last weight adjustment: October 3, 2026

Tiers

S

Score 85 or higher

A

Score 70 to under 85

B

Score 55 to under 70

C

Score under 55

When two displayed scores are equal, the unrounded score decides the order, then the name.

How we curate (not a full catalog)

• We refresh research weekly, and sooner when something important ships. We don't try to list every tool: each task list is curated, and entries are pruned when both recency and reputation fade.
• New candidates are expected to show credible activity within roughly the last six months (releases, changelogs, or pricing updates).
• Older entries can remain when they still show strong reputation signals (e.g. sustained community adoption) even if they are not brand-new.
• We may flag entries for review when both recency and reputation signals look weak; that is guidance for editors, not an automatic delist.
• A task ranking includes only tools whose main use is that task, ordered by score. Tools that also fit the task are listed below it as "Also used for", without a rank. A tool's page shows its rank within its main task.

Limits we acknowledge

• No ranking is fully objective — benchmarks do not capture all real-world use.
• AI-collected community sentiment can contain errors; we verify before publishing.
• We represent an opinion grounded in public data — not authoritative judgment.

We don't publish evaluations without verifiable sources.

Corrections

Found a mistake? Report incorrect information — product owners get a reply within 24 hours. Every ranking change we publish is listed in the changelog.