The World's Most Capable AI Model Is Now Standard on Max — at Half the Limit

The World's Most Capable AI Model Is Now Standard on Max — at Half the Limit

On July 2 we wrote the story of Fable 5: the model Anthropic launched to the top of every leaderboard, that Washington pulled three days later, and that came back after 18 days. What wasn't clear then was the practical question: it's back — but for whom, and on what terms?

Today, July 20, 2026, we have the answer. The temporary period where everyone got to try the model has ended, and the rules are now permanent. The short version: access to the most capable model is no longer a question of "do you subscribe?" but of "which tier are you on?"

What changed today

Per Anthropic's official support page, the promotional period ended the evening of July 19 and the new rules took effect on July 20. The split looks like this:

• Max plans and premium seats on Team and Enterprise plans: the model is included in your subscription, and you can spend up to 50% of your weekly usage limits on it at no extra charge.
• Pro plans and standard Team seats: not included in your plan's limits. You use it with paid usage credits, and eligible accounts get a one-time credit to ease the transition.
• Standard Enterprise seats: also not included, and your organization must have usage credits enabled. These seats don't qualify for the promotional credit.
• API usage: billed at standard API rates, as always.

One detail that matters if you're on Max: hitting the 50% mark doesn't stop your work. You get two clear options — keep going on Fable 5 with paid usage credits, or switch to another model and keep working inside your plan's normal limits.

Why half the limit, not the whole thing

Because Fable 5 is expensive to run in a different league. Anthropic says plainly in its own messaging that the model draws down your usage faster than other models. That's not a footnote: an hour of work on Fable 5 is not equivalent to an hour on a lighter model, even though the meter looks the same.

In practice, half a weekly limit on a hungry model is a lot less than it sounds on paper. Anyone who reaches for the strongest model for every small task will hit the ceiling mid-week.

The bigger signal: frontier AI is becoming metered

A flat monthly subscription implied that intelligent compute was a cheap, effectively unlimited commodity. What's happening now says the opposite: models at the frontier are being priced with the logic of electricity and fuel rather than of an app subscription. You pay for what you consume, and the consumption differences between models are real.

This isn't only Anthropic. The same logic is showing up across providers: a capped, expensive top tier and cheaper tiers for daily work. Anyone planning AI usage inside a company needs to budget on that basis, not on the assumption that the strongest model is available without counting.

What to actually do if you run a business

1. Sort tasks by real difficulty. Most daily work — summarizing, drafting, classifying, replying to a customer — does not need the strongest model. Reserve the top tier for problems where the tier below it actually failed.
2. Know where your usage goes. If you hit the ceiling every week, the problem is usually task allocation, not plan size. Find out where the limit drains before you upgrade.
3. Don't build a critical process on a single tier. If your operation stops when one model's limit runs out, you're exposed. Make the system able to fall back to a lighter model without halting the work.
4. Cost per task, not per month. The right question isn't what the subscription costs — it's what a single task costs and what it saves in return.