Claude Fable usage limit: how the 50% weekly cap works

What "up to 50% of your weekly limit" means

It's a cap on Fable, not a bonus. Anthropic's help page on Fable models says Max and Ultra subscribers can use "up to 50% of your weekly usage limits on Fable models at no extra cost", and that Fable draws "from your plan's regular weekly usage limits".

So there are two meters. Your normal weekly limit covers every model. A second, separate weekly limit stops Fable once it has eaten half of that total. Anthropic's answer to whether this means 50% more usage is short: "No." You can never use more than your weekly limit.

In practice: if you spend half your week on Fable, Opus, Sonnet and Haiku get the other half. If you never touch Fable, the whole limit goes to the other models. If you use Opus all week first, there may be less than half left for Fable by the time you switch.

When does the Fable limit reset?

Once a week. Anthropic says on its usage credits page that Max and Ultra have "a separate weekly limit for Fable that resets once a week", and its Max plan page says weekly limits reset at a fixed day and time assigned to your account. It stays the same no matter when you started using Claude or when your subscription began.

Both meters are in Settings, then Usage, along with your next reset time. Ultra accounts that are verified have no five-hour limit, so on Ultra the weekly meters are the only ones that matter. See Claude limits explained for the full plan table.

Why Fable drains your limit faster

Anthropic says Fable uses your limits "faster than other Claude models". It doesn't say how much faster. The API price is the best public hint: Fable 5.1 costs $10 in and $50 out per 1M tokens, against $4 and $20 for Opus 5.5, per the Claude pricing page. That's 2.5x on output.

If plan usage tracks API cost (our assumption, not Anthropic's statement), an hour of Fable takes about as much of your week as 2.5 hours of Opus. Long agent runs in Claude Code are where this bites, because they pile up output tokens.

What happens when you hit the Fable cap

You get two choices, per Anthropic: keep using Fable with usage credits, or switch to another Claude model and stay inside your plan. Switching costs nothing. Opus 5.5 is the natural step down and the rest of your weekly limit is still there for it.

Usage credits are billed at standard API rates, per the usage credits page. You turn them on in Settings, then Usage, prepay a balance and can set a monthly spend limit. They don't change when any limit resets. For our sample coding session (2M fresh input, 8M cached input, 0.5M output tokens), Fable 5.1 on credits costs about $47.00, Opus 5.5 about $19.60 and Sonnet 5.5 about $9.80 (our estimate at list price). More on this in Claude extra usage.

Can you use Fable on Claude Pro?

Yes, but only on usage credits. The pricing page lists Pro with "Fable, paid with usage credits", and Fable doesn't count against Pro's plan limits at all. Free has no Fable.

This changed in July. A promotion that let paid plans use up to 50% of the weekly limit on Fable 5 ended on July 19, 2026 at 11:59:59 PM PT. Eligible Pro and Team standard seats got a one-time credit for that change. Anthropic says there's no equivalent credit for Fable 5.1, which was never part of the promotion.

Here's the math for Pro users who like Fable. The $80 a month between Pro and Max buys about 1.7 of our sample sessions on Fable credits (our estimate). If you'd use Fable more than that, Max is cheaper, and you get 5x Pro's usage on the other models too. See is Claude Max worth it.

Fable vs Opus 5.5: which should you spend your week on?

For coding, Opus 5.5. It ranks #1 of 50 on LMArena's Code Arena against Fable 5.1 at #5 of 50, and it costs less per token. On a plan, that means more coding per week for a model that ranks higher.

Fable earns its place on hard reasoning and long documents. It ranks #2 of 51 for reasoning and math (Opus 5.5 is #5 of 51) and #1 of 11 on the Document Arena, which has no Opus 5.5 score yet. A sensible split: Opus for the daily coding grind, Fable for the problem Opus keeps getting wrong. The numbers side by side are in Claude Opus 5.5 vs Claude Fable 5.1.

Why your Fable chat switched to Opus

That's a safety fallback, not a limit. Fable runs classifiers on every request, and some topics (offensive security, dual-use biology, a narrow set of AI chip and training work) get rerun on an Opus model, per Anthropic's model switching page. The picker then stays on Opus for that conversation.

The Opus answer is charged separately at Opus rates. You can turn this off with the "Switch models when a message is flagged" setting under Settings, then Capabilities, in which case a flagged message pauses the chat instead.

Using Fable in Claude Code

Fable 5.1 needs Claude Code version 2.1.255 or later, per Anthropic. It shares the same weekly limits as chat, so a long Fable session in the terminal uses the same 50% cap. If your week keeps running out, see the Claude Code weekly limit and the cheapest API fallback when you hit it.

Related pages

• Claude Fable 5.1 API pricing and Claude Opus 5.5 API pricing.
• GPT-6 Astra vs Claude Fable 5.1 and Claude Opus 5.5 vs Claude Sonnet 5.5.
• Best models for reasoning and best models for long documents.