Claude's weekly limit: all models vs Fable

On a Max plan and still ran out of Claude? Max runs two separate weekly caps - one across every model, one scoped to Fable - and they produce completely different symptoms. One blocks everything; the other just moves Fable onto pay-as-you-go credits. Pro runs a single weekly cap across all models. Once you know which cap you hit, the path forward is obvious.

Update, 2026-09-19: The separate weekly Opus budget that Max plans used to run no longer appears. Opus and Sonnet are both current, fully available models, and both draw from the single weekly limit that covers every model. The second weekly meter on Max is scoped to Fable instead. Checked against a live Claude Max 5x account that used Fable, Opus and Sonnet in the same week: Settings > Usage shows exactly three meters - current session, this week, and Fable this week - and no Opus meter. Sections below that describe an Opus-only weekly cap are kept as history and marked "(Retired: ...)", because that is still what people search for.

Update, 2026-09-18: Anthropic's Fable models (Fable 5, and Fable 5.1 since 2026-09-01) sit above Opus. On Max plans they are included for up to 50% of your weekly limit at no extra cost, and they drain that limit faster than Opus or Sonnet; past the 50% mark Fable keeps running on pay-as-you-go usage credits. On Pro, Fable is not part of the plan limits at all: every Fable message is billed from usage credits. The Free plan has no Fable access. The weekly windows described below work the same way; on Max, Fable simply spends from the same weekly allowance, faster.

The short answer

Claude's rolling limits: two on Pro, three on Max

Most users know about the 5-hour session cap - it is the one you hit mid-session that makes you wait a few hours. What catches people by surprise are the weekly caps, especially on Max, where there are two of them running on different scopes and producing different user-visible effects.

Per Anthropic's usage limits documentation, the three caps are:

Retired: Max plans used to run a third weekly cap scoped to Opus only, which silently downgraded you to Sonnet when it ran out. That meter no longer appears on the usage screen, and Opus usage counts against the all-models weekly cap like every other model. Opus itself is unchanged and fully available.

All three windows are rolling - they do not reset at midnight, on Monday, or at the start of the billing month. Each starts from your first relevant message of that particular cycle. This is covered in full in how Claude's usage limits really work in 2026 and confirmed by the community thread on Hacker News discussing why Claude's weekly limits don't reset every 7 days on a fixed schedule.

The two "I'm on Max and still ran out" scenarios

Here is the exact confusion that brings most Max plan users to this page. You pay $100 or $200 per month for a Max plan specifically to get more Claude capacity. Then you hit a wall anyway. Two very different things could be happening:

Scenario A - Claude is serving you Sonnet instead of Opus (retired cause)

You open a new chat, select Claude Opus, and send a message. The response comes back quickly but feels less capable than usual. Or you notice the model indicator in the UI quietly shows Sonnet rather than Opus. Or the responses on complex reasoning tasks are thinner than you expected.

(Retired: this used to be the Opus-specific weekly cap.) Max plans historically ran a separate weekly Opus budget, and when it ran out Claude silently downgraded you to Sonnet with no error message. That meter no longer appears, and Opus draws from the all-models weekly pool along with every other model.

If you are seeing Sonnet answers today when you asked for Opus, check the model picker first - it can fall back to Sonnet in a new chat, and Claude Code can be configured to choose the model for you. Then check Settings > Usage: on Max you get three meters (current session, this week, Fable this week), and the weekly meter covers Sonnet and Opus together.

Scenario B - Everything is blocked

You try Sonnet. You try Haiku. Nothing works. Claude shows a rate-limit or usage-limit message with a reset timestamp that is one to seven days in the future. This is the all-models weekly cap. Every model on your account is locked until the rolling 7-day window resets - even on Max 20x. This is a hard ceiling, not a soft downgrade.

Hitting the all-models weekly cap takes sustained heavy use, especially on Max 20x. But it is entirely possible if you run Claude Code heavily all week on large codebases, or if you drive many long agentic sessions. Anthropic does not publish the absolute budget numbers, so the only way to know how close you are is Settings > Usage on claude.ai, which shows the percentage consumed for each cap.

How to diagnose which cap you hit

The fastest diagnostic is a two-step test:

1. Open a new chat and send a simple message. If it goes through on any model, your all-models weekly cap is still open - Sonnet, Haiku and Opus all draw from that one pool, so if one answers, none of them is capped.
2. If Sonnet is also blocked, check the error message or the reset timestamp. A timestamp days away means you have hit the all-models weekly cap. If the timestamp is hours away, you may have hit the 5-hour session cap instead (much more common).

For a definitive reading, go to Settings > Usage on claude.ai. It displays separate progress meters for your current session, your weekly all-models budget, and (on Max plans) your Fable weekly meter. Each meter shows the percentage consumed and the reset timestamp. In Claude Code, run /usage in your terminal for the same data.

For a live, ambient view without opening any settings page, tokn.watch shows your session and weekly meters in your Mac menu bar as colored arcs that update in real time.

The rolling window - what "7 days" actually means

Claude's weekly caps use rolling 7-day windows, not calendar weeks. This is one of the most commonly misunderstood facts about Claude's limits.

The weekly all-models window starts when you send your first message of the cycle. If you sent that message on a Tuesday at 14:00, your weekly reset happens the following Tuesday at 14:00 - not Sunday night, not Monday morning. On Max, Fable does not get a second clock: its 50% share is a ceiling inside that same weekly window, so Fable can never take you past your weekly limit.

The practical consequence: if you exhaust your weekly pool on a Thursday, you do not get it back Monday - you wait until the same time next Thursday. This is what the Hacker News thread on Claude's weekly limits documents in detail: users who expected a Monday reset discovered their window does not line up with any calendar boundary.

This also means your effective "weekly budget" per calendar month is not four equal blocks. Depending on when your cycle started, the number of complete windows per month varies by one. Heavy users who plan their Claude Code work around weekly capacity need to track the rolling timestamp, not assume Monday is the reset day.

How Max 5x and Max 20x change the budget

Max 5x gives approximately five times the budget of Pro across all three caps. Max 20x gives approximately twenty times. These are the multipliers Anthropic has publicly described; they do not publish the absolute usage-unit numbers for any plan.

What this means in practice:

• A Max 20x user doing the same weekly workload as a Pro user will hit the all-models cap roughly 20 times less often.
• Fable's 50% share is proportionally larger on Max 20x than on Max 5x, because it is half of a bigger weekly number.
• But neither Max plan is unlimited. Very heavy Claude Code users - especially those running long agentic sessions on large codebases all week - can exhaust even Max 20x before the 7-day window closes.

According to Anthropic's usage limit best practices, the best way to stretch any plan's weekly budget is to match the model to the task - use Haiku for fast lookups, Sonnet for most coding work, and reserve Opus for tasks that genuinely require its reasoning depth. On Max, Fable drains the weekly limit fastest of all, so saving it for the tasks that need it stretches the week the furthest.

"Claude's usage limits are designed to provide fair access across all users. We recommend using the most appropriate model for each task rather than defaulting to the most capable model for every interaction." Anthropic - usage limit best practices (paraphrased)

The silent Opus downgrade (retired)

Retired. Max plans used to run a weekly budget scoped to Opus only, and when it ran out Claude quietly answered as Sonnet instead. That meter no longer appears on the usage screen, and Opus draws from the all-models weekly pool now. This section is kept because it is still what people search for. Here is how it worked while it existed:

• No hard error message. You were not shown "Opus is at limit." Claude simply started responding as Sonnet. The model indicator in the UI updated if you looked, but there was no proactive notification.
• Sonnet quality for Opus tasks. For most coding and writing tasks, Sonnet is nearly as good. For multi-step reasoning, long document analysis, or highly creative work, the quality gap is noticeable.
• Your overall weekly budget kept depleting. Even in Sonnet fallback mode, the all-models weekly cap carried on ticking down. Sonnet is more efficient than Opus per task, so the weekly pool lasted longer - but it was never paused.
• The Opus cap ran on its own timer. The Opus window started from your first Opus message of the Opus cycle, not your first message of the all-models cycle. They were independent clocks. Nothing on the usage screen works that way now.

For Claude Code users: /usage shows whichever meters your account actually has - the session and weekly pools, plus the Fable meter on Max. Checking it before starting a heavy agentic task tells you whether you have enough weekly budget to finish.

How the session cap fits in

The 5-hour session cap is independent of the weekly caps. You can hit the session cap multiple times in a single week - each time you just wait a few hours before the session pool refills. It does not consume any of your weekly all-models budget when it is locked; the weekly clock keeps running, but the session cap is its own separate lockout.

The most confusing scenario: you hit the 5-hour session cap on a Friday afternoon, then the weekly all-models cap on a Friday evening. The session cap resets by midnight (5 hours from first message), but the weekly cap is still locked. You can send a message after the session reset, only to be told you are still at your weekly limit. The two caps compound independently.

A full comparison of all three caps and how they stack is in the deep dive on Claude's 5-hour session limit.

What to do when you hit each cap

If Claude is answering as Sonnet when you wanted Opus (Scenario A)

Set the model explicitly rather than relying on the default, and check Settings > Usage. There is no Opus window left to wait out: if the weekly meter still has headroom, Opus is available and the model picker is what needs changing. If the weekly meter is empty, no model will run until the rolling 7-day window resets.

For future prevention: save Opus for tasks that genuinely need it. Use the model picker to default to Sonnet for exploratory coding, file reading, and drafting. Reserve Opus for the final reasoning pass on complex tasks. Because everything now shares one weekly pool, that choice stretches the whole week rather than one model's budget.

If you hit the all-models weekly cap (Scenario B)

You have no Claude access until the rolling 7-day window resets. Check the reset timestamp in the error message or in Settings > Usage. If you need to keep working right now, the practical alternatives are Claude.ai's free tier (which runs on a separate pool), ChatGPT (different provider, completely independent limits), or a direct API call if you have an Anthropic API key with a separate billing quota.

For future prevention: track your weekly budget visually so you can see it depleting before it hits zero. The week-level budget runs on a 7-day clock - if you are at 80% by Wednesday, you are on track to exhaust it before the week-cycle ends. Slowing down or switching to lighter models mid-week preserves more capacity for time-sensitive tasks later in the week. The tactics in 13 ways to save Claude tokens apply equally at the weekly level.