Live events & highlights

Temporary promos/throttles with countdowns, plus the big stuff: model upgrades, price moves, new plans. The exhaustive per-limit log lives in Changelog.

Radar

Auto-flagged news (unverified)

Items the discovery bot pulled from provider feeds and flagged as limit/pricing-relevant. Machine-flagged, not verified — confirmed changes get promoted to What's new with a sourced quote. auto

Sortable list By provider By price

Current snapshot

Latest published limits

One row per plan, native units as published. Group by provider or by price tier. We do not invent message counts where providers only give token budgets or multipliers. ? means honestly unpublished. Best value per $ ->

How to read this

Confidence — official provider page · announced blog/exec post · community 3rd-party measured · crowdsourced user reports

Sources — values with a dotted underline are quotes: hover to read the exact source sentence, click to pin it. Every row also links its src.

Units — messages, tokens, and requests are absolute caps. 5× Pro is a relative multiplier with no published absolute. $20 usage is a monthly dollar credit, not a throughput cap.

Windows — /3h /4h /5h /day /wk /mo is the refresh period. ? means the provider hasn't published a number.

Hard-cap math

Value per dollar (theoretical ceiling)

Throughput per $ if you used the whole limit every window. A ceiling, not typical use. Compared only within the same unit. We never silently equate a message to a token. Multiplier-only plans (no published absolute) show as relative and can't be valued.

Best cross-provider view

Cross-provider estimate (tokens / day / $)

This is the most useful comparison, but it has assumptions. Message caps are converted to tokens, dollar credits are converted using a token price, and requests-only plans stay out because a request is not a token.

Tune the assumptions below. The ranking updates live, and the table labels estimated rows instead of pretending they are hard facts.

tokens per message: 700 credit $ / million tokens: 5

Provider / Plan	$/mo	est. tokens / day / $

Real receipts

What plans actually delivered (real accounts)

Observed consumption from maxed-out accounts, for plans that publish only a dollar credit. This is not a cap; it is what a real billing month delivered in tokens. crowdsourced

Field reports

Real usage observations (read off the usage screen)

What real accounts actually hit on chat subscriptions, read straight from the provider's own usage screen (claude.ai, ChatGPT, Claude Code). These caps are not in any API, so this is the only way to measure them. Each row is one observed reading, kept verbatim. crowdsourced

Submit a reading →

All individual readings

Captured	Plan	Window	Observed	API value	Note

Drift tracker

Price history

Each plan's published price across every snapshot we have. Backfilled from the Wayback Machine, so the line grows as history accumulates. Hover a sparkline for the dated points.

Drift tracker

Limit changelog

Derived by diffing snapshots. This is the part nobody else tracks: when limits silently move.

Provenance

Why you can trust this

Every number carries a confidence tier and a link to its source. Hover any dotted limit to read the exact source sentence.
We store only as-published facts. Per-day / $-value / cross-provider math is derived in your browser, never baked into the data.
Raw dated snapshots are public and version-controlled. The git history is the audit trail.