A lot of talk about reduced weekly limits but a lack of data - so here's some actual numbers

Two questions answered by actual data, not just vibes: • How much the Claude Max weekly limit actually dropped after the promo ended • Max 5x vs Max 20x and what that actually means for usage limits There's a lot of "my limits got slashed way more than they said" going around with no numbers behind it. So here are some actual numbers, from my own account. TL;DR The weekly-limit cut is real and roughly the size Anthropic announced. Weekly limits in Claude Code were +50% from 13 May to 13 Sep; from 14 Sep they are +25%. That's an announced cut of 17% (150% → 125% of normal). I measure 19% (150% → ~122%), with a plausible range of 10–25%. It took effect at the weekly reset, not on 13 Sep. My weekly reading sat at 30% straight across the 13th with no jump, and 13–14 Sep burned at the old rate. Max 20x is not 4× of Max 5x. Same account, July on 5x → August on 20x, same models: the 5-hour session limit went up 4.4× , the weekly limit only 2.2× . On 5x the weekly bucket holds ~11 full sessions; on 20x only ~5. Rough sizes , in Claude Code work valued at Anthropic's API list prices: Max 5x ≈ $790/week and $70 per 5-hour session ; Max 20x ≈ $1,710/week and $310 per session (Fable 5 era, promo +50% weekly limit active). If you moved to Fable 5.1 around the same time, your quota probably "halved" for a different reason. See the footnote. Setup One Claude Max account — Max 5x , except all of August on Max 20x . Every Claude Code session transcript is parsed into a SQLite ledger (tokens per model per hour, cache split), and the same /usage endpoint the CLI reads is polled every 2 minutes, logging both the 5-hour and 7-day utilization. Tooling: github.com/eksdeexD/ccstats Method I log how much Claude Code work, valued at API list prices, it takes to move each limit by one percentage point. Fewer dollars per point = smaller limit. Two things could distort that: usage outside Claude Code (claude.ai chats, ~8% of my usage this week) burns the limits but isn't in my transcripts, and I don't know Anthropic's exact token weighting. So I use the 5-hour session limit as a reference : it was never part of the promo (the support article says so explicitly), and chat usage hits both buckets the same way. Comparing the weekly limit against the session limit cancels both distortions out. Sanity check: that weekly-to-session ratio came out 11.2 on 5x in July and 11.1 on 5x in September — different models, two months apart. Excluded from the data: the two days where plan switched, every period where extra usage was being consumed, and one mid-week free reset Anthropic handed out. Numbers Period Plan Weekly limit ≈ 5-hour limit ≈ Weekly ÷ 5-hour Jul 10–30 (promo +50%, Fable 5 / Opus 5) Max 5x $791 $70 11.2 Aug 1–30 (promo +50%, Fable 5 / Opus 5) Max 20x $1,712 $312 5.5 Sep 5–14 (promo +50%, Fable 5.1) Max 5x $421 $38 11.1 Sep 15 → (new +25%, Fable 5.1) Max 5x $469 $52 9.0 Weekly cut: ratio 11.1 → 9.0, so the new weekly limit is ~81% of the promo one, i.e. down by 19% (announced as 17%). Relative to the normal, pre-promo limit: 150% → ~122% (announced 125%). Two days / 29 weekly points on the new side so far, so read it as down by 10–25%. I'll update as the week fills in. 5x vs 20x (July → August, same models): weekly 2.2× larger, session 4.4× larger. "20x" describes the session limit; the weekly and monthly limits are roughly double, so a more honest name would be "Max 10x" Don't compare the dollar figures across the model change (July/August vs September) — see the footnote for why. Footnote: the Fable 5.1 effect (why your "API-equivalent" quota looks halved) On the same 5x plan, my September usage (Fable 5.1) shows roughly half the API-list-price value per weekly point that July (Fable 5) did. Per token it's nearly the same (~1.1–1.3×). The difference is one line item: Fable 5.1's cache reads are listed at $0.25/MTok versus $1.00 for Fable 5, and ~97-99% of a Claude Code session's tokens are cache reads. The plan limits don't appear to pass that discount through — from the limit's point of view a 5.1 token costs the same as a 5 token. So if you switched to 5.1 around the promo end and your cost-tracker numbers say your weekly quota halved, this is very likely why, not the promo. (Inference from the data, not something I've seen Anthropic state.) Promo terms: support.claude.com article 15910845 (Claude Code weekly limits +50% May 13–Sep 13, +25% permanent from Sep 14; 5-hour limits unaffected). Readings are integer percentages, so single low-usage days are noisy; the figures above pool multiple days. This post was - you guessed it - written by Fable 5.1 and slightly improved by me. If anything is unclear or doesn't read easy, let me know.