Methodology
We read the official pricing page, copy what it says, link it, and date it.
Sources
Each row cites one primary source, the vendor's own pricing page, and may list supporting official docs (shown in the row's expanded view). Right now the dataset has 118 rows across 29 coding tools, citing 72 distinct URLs. Values are copied as written, not paraphrased, and never rounded.
Status labels
- ✓ Verified: every field in the row was read from a cited source on the check date. (81 rows)
- ◐ Partial: at least one field couldn't be confirmed. That field shows Not confirmed. (37 rows)
- ! Stale: not re-checked in over 30 days. The freshness dot turns amber after 7 days and red after 30.
Status is always shown as an icon plus a word, never color alone.
What "limit" means, by billing model
Units aren't comparable across tools, so we don't publish a "best value" score.
- Credits / tokens (Lovable, Bolt.new, v0): A pool of units consumed per action. Cost per prompt varies by task, so we never convert credits into "prompts".
- AI credits (dollar-denominated) (GitHub Copilot, Replit, v0): An included dollar amount spent at per-model rates. We record the dollar figure as stated.
- Rolling window + weekly cap (Claude Code): Usage refills on a rolling session window with an additional weekly ceiling. Cap sizes are not published as numbers.
- Shared weekly pool, multiples (Cursor): A usage pool refreshing weekly; higher plans are stated as multiples of a base plan (e.g. 4×). We show the multiple, never a %.
- Unquantified tiers (Windsurf): Quotas described in words only. We quote the words.
Derived fields
A few display aids are computed at build time, never stored as facts: the reset chip (keywords in the reset text), the overage icon (keywords in the overage text), the price band (first dollar figure in the price string; unparseable prices go in "Custom"), and the tier ladder (a plan's position in its own tool's lineup). Every count on this site is computed from the data at build time.
Cadence
The plan is weekly re-checks of every pricing page, plus ad-hoc checks when a vendor announces a change. Each change is logged in the changelog with its source. The latest check date in the data is .
Model scores, use cases and costs
Scores are third-party leaderboard results copied as published, each with its board, metric, harness, as-of date and source link. We never combine, normalise or weight them, and there is no Ratelimit Now score. Results older than 90 days carry a Stale chip; boards that stopped scoring new models carry Frozen; Preliminary is the board's own label. iOS has no mature public leaderboard, so it shows none.
Use cases link to the builder's public post. Every link is re-fetched before a build, and an item whose post is gone or doesn't match is dropped. Vendor, ambassador, employee, self-promo and second-hand posts are tagged.
Build cost is spend that builders state in public posts (X, Reddit, HN, blogs, dev.to), labelled "Estimate from user reports". Per category and tool we show p25–p75, median and min–max only when n is 3 or more; otherwise the reports are listed one by one. Paid spend and API-equivalent spend are kept apart. Vendor claims, promo, second-hand posts, upper bounds and numbers seen only in a search summary are listed but not counted. See build cost.
Report outdated info
There's no correction form yet. Until there is, send the page link and a source to @QuietlyByG on X. Every accepted correction appears in the changelog with its source.
Independence rules
- Sponsors never change data, ordering or sorting.
- No paid placements inside the tracker table.
- Any affiliate link is labelled "Affiliate link" in plain text. This build contains none.
- Vendor sponsorship of the tracker is off by default. See how we make money.