Anthropic resets Claude Max limits for Fable 5.1

Anthropic's @ClaudeDevs account posted on September 1, 2026 that Claude Fable 5.1 was live in Claude Code and on the Claude Platform — and, in the same beat, that Anthropic had reset 5-hour and weekly limits for all users. Max subscribers opened usage settings and saw meters at zero. That flush was timed with the model launch, not a change to the Max plan's Fable cap. The help article on Fable models on your plan still says Max, premium Team seats, and premium seat-based Enterprise seats may spend at most 50 percent of the weekly limit on Fable 5 and Fable 5.1, drawn from the same pool as every other Claude model.

This briefing is for Max 5x and Max 20x subscribers, Claude Code users on 2.1.255 or later, and anyone routing Fable 5.1 through the API. It covers what Anthropic actually reset, how the 50 percent Fable bucket works, why Fable 5.1 still burns a Max week faster than Fable 5, who is not included, and what to verify before you treat a zeroed meter as extra headroom.

What happened

The public notice is the September 1 @ClaudeDevs post: Fable 5.1 is in the Claude Code model picker as claude-fable-5-1 on the platform, it is priced the same as Fable 5, and API cache reads are 75 percent cheaper. The same thread said 5-hour and weekly limits were reset for all users. Developers posted screenshots of empty weekly bars and started weekend agentic sessions against the new model. That is a usage-counter flush, not a new Max SKU and not a permanent lift of the Fable share.

Anthropic's help center is explicit that the July 19, 2026 promotion — Fable 5 inside weekly limits for more plan types — ended at 11:59:59 PM PT that day and applied to Fable 5 only. Fable 5.1 was never part of that promo. On Max, both Fable 5 and Fable 5.1 are a standard part of the plan, subject to the 50 percent weekly Fable ceiling. On Pro and standard Team seats, both models run on usage credits from the first token. There is no one-time credit for Fable 5.1 matching the July Fable 5 credit.

How it works

On Max, Fable draws from the regular weekly usage limit and from a Fable-specific share of that limit. The help center FAQ answers the misread directly: you do not get 50 percent extra. You can use up to 50 percent of the weekly limit on Fable models; other models draw from the same weekly limit; you never exceed the weekly cap. Fable also uses those limits faster than other Claude models. When the Fable share is gone, Max users either keep going on usage credits or switch to another Claude model and stay inside the remaining weekly budget.

Fable 5.1 itself changes the burn rate even when the meter starts at zero. GitHub issue 91623 on anthropics/claude-code records one Max 20x account going from a reset counter to 88 percent of the weekly limit in about 22 hours after the 5.1 launch, with the Fable bucket ahead of the overall weekly meter. In five long-running sessions that used both models on that machine, Fable 5 averaged 783 output tokens per turn and Fable 5.1 averaged 1,596 — about 2.04 times. Across every session on that machine the ratio was 2.07 times. Artificial Analysis, in pre-release testing, measured about 1.7 times the output tokens of Fable 5 at max effort. Effort is the control: Anthropic's Fable 5.1 prompting guidance says to start at the default high, then re-sweep levels because names do not match Fable 5 thinking. Public notes from launch day say Fable 5.1 at medium is meant to land near Fable 5 quality at lower cost, and Lance Martin's September 1 tips reported Fable 5.1 at low effort at parity with Fable 5 at high effort on CursorBench 3.2.0 at about a third of the cost. Mid-conversation effort changes can keep the prompt cache if you use the documented beta path; a naive model or effort switch still invalidates it.

Why it matters

A reset is a one-time zero. The 50 percent Fable rule and the faster Fable burn are standing policy. If you spend the flushed week on Fable 5.1 at high or max effort in Claude Code, you can empty the Fable bucket and then the weekly meter before the next weekly cliff. Prompt-cache economics cut the other way: Fable 5.1 cache hits are billed at $0.25 per million tokens versus $1.00 on Fable 5, which is the 75 percent cut @ClaudeDevs cited. That saving only shows up if the cache stays warm. Cold cache after a 5-hour lockout, a compact, or an effort switch rewrites the prefix at full input rates and can take a large bite of a 5-hour window on a Max 5x plan — Reddit reports of a single continue on an 800K-context project eating around 20 percent of the 5-hour limit are consistent with a cache miss, not a billing bug.

For agentic coding, that mix — reset, 2x-ish output per turn, 50 percent Fable share, cheaper cache reads — is why screenshots of zeroed meters and GitHub threads about a drained Max 20x showed up in the same 48 hours. The extra room is real for a few days. It is not a license to leave Fable 5.1 on max effort across every session. Keep Sonnet or Opus as the compact-and-continue path when the 5-hour window is about to close, then return to Fable 5.1 on a warm cache if you still have Fable share left.

Who is affected

Max 5x, Max 20x, premium Team seats, and premium seat-based Enterprise seats are the plans that include Fable 5.1 in weekly limits at the 50 percent cap. They saw the September 1 flush. Claude Code needs version 2.1.255 or later to select Fable 5.1; Fable 5 needed 2.1.170. Cowork needs current Claude Desktop. Pro, standard Team seats, and standard Enterprise seats still pay usage credits for Fable 5.1 with no matching promotional credit. Free accounts do not get Fable at all.

API and usage-based Enterprise customers are on published API rates, not the Max weekly meter, so the counter reset does not change their invoice. They do get the Fable 5.1 cache-read price of $0.25 per million tokens and the same effort controls. Builders who already sued or complained about Max usage advertising should read the help-center language, not the launch tweet, as the contract-shaped description of the 50 percent cap.

What to watch next

On August 29, @ClaudeDevs said that starting September 14, Anthropic would permanently raise standard weekly limits in Claude Code by 25 percent for Pro, Max, Team, and seat-based Enterprise plans. That is a separate change from the September 1 flush. Confirm it on the usage page that week rather than assuming the reset already included it. Watch your own usage settings for both the weekly bar and the Fable bar; the help center says you can track both.

Verify the model string claude-fable-5-1, Claude Code 2.1.255+, and whether your org has Fable enabled. Re-run effort sweeps on Fable 5.1 instead of copying Fable 5 defaults. If a Max 20x week still evaporates in a day, log output tokens per turn and cache hit rate before you file a metering bug — issue 91623 is the public comparison set. Do not change production routing on a zeroed meter alone; the cap and the burn rate are still in the help article at support.claude.com.