ToolScore

Methodology

Five-star ratings are broken — they hide how they were formed. The ToolScore Index is 10 criteria × 10 points, each score visible, each number sourced and dated. If a tool scores 61/100, you can see exactly which pillars cost it points.

The 10 criteria

CriterionWhat 10/10 looks likeWhat 0–3 looks like
Price / ValueFree or cheap with real capability; predictable cost$40+ tiers, metered surprises, teaser entry price
Token & Rate LimitsA plan holds a full working day of heavy use$20 dies in 2–3 hours of active work
Speed / LatencyStreaming, responsive, interactiveMulti-second stalls, queueing
Privacy & RetentionNo training on your code by default; zero-retention or self-hostTraining by default, opt-out buried
BYOKBring your own API key; pay raw token pricesLocked to vendor's model menu and markup
Output QualityTop-tier benchmarks and real-world task closurePlausible-but-wrong code, weak reasoning
StabilityPredictable behavior, uptime, few regressionsBroken sessions, breaking changes
Ease of SetupMinutes from install to first real taskConfig archaeology before productivity
Billing TransparencyYou can predict the invoice before it arrivesFuzzy windows, credit meters, moving caps
Support & CommunityDocs, responsive channels, active ecosystemCommunity-only, stale docs

Verdicts: 80+ Best Value · 65+ Strong · 50+ Average · <50 Weak.

Where the data comes from

Three layers, always labeled on the page:

Official sources — pricing pages, docs, changelogs. Dated because they change.
Community signals — Reddit, Hacker News, GitHub issues: where the real pain points live. Labeled [community] and never presented as our own testing.
Internal tests — our own measurements, with method described. Labeled with date and link here.

Fact verification

Every price, limit, and policy claim we publish goes through an automated verification pipeline: the claim is extracted, re-searched against independent sources, and marked verified only when an authoritative source confirms it. Claims that can't be confirmed stay labeled unverified or get cut.

A site that debunks hollow data can't publish unverified numbers as facts — that constraint shapes everything here. If you find a stale number, report it; we'd rather correct than defend.

Scores vs. affiliate links

ℹ️ The firewall
Affiliate relationships never touch scores. Tools without any affiliate program regularly top our rankings — that's the proof. Money details live in the disclosure; the scoring pipeline has no knowledge of them.

Freshness

Every page carries a “data as of” date. Pricing and limits in this category change monthly; anything older than ~60 days without a refresh should be treated as suspect — including by us. Our pipeline re-checks on a rolling basis and re-scores when the facts move.