On Friday July 18, 2026, Anthropic told its subscribers what happens on Monday. Beginning July 20, Claude Fable 5 — the premium frontier model that spent nineteen days offline under a Commerce Department export-control directive last month and came back on July 1 — becomes a permanent feature of the Max and Team Premium plans, at 50 per cent of their weekly usage limits. On the same day, those weekly limits themselves drop by one third, as the fifty-per-cent bonus phase Anthropic ran through the redeployment window expires. Pro and Team Standard subscribers lose Fable 5 from the flat monthly fee, receive a one-time $100 usage credit, and after that pay API rates. Read the sentence twice: half of a smaller cap, for the top tier, and a two-hundred-dollar off-ramp for everyone below.

The move was carried in a thread on Anthropic's X account and picked up within hours by The Decoder, Simon Willison, DAWN, TechTimes, ForkLog and AI Pricing Guru. The last of those republished the API rate card — $10 per million input tokens, $1 per million cached input, $50 per million output — the same schedule Anthropic set at the July 1 relaunch and did not touch on Friday. What changed on Friday is which subscribers hit that rate card first.

What each tier gets on Monday

The plan-by-plan grid, drawn from the Anthropic thread and the AI Pricing Guru breakdown:

• Pro ($20/month): loses Fable 5 from the flat rate. Anthropic issues a one-time $100 usage credit that expires when it is spent. After that, Fable 5 access rolls onto pay-as-you-go at API rates.
• Max 5x ($100/month): Fable 5 included, at 50 per cent of the weekly limit.
• Max 20x ($200/month): Fable 5 included, at 50 per cent of the weekly limit.
• Team Standard: same treatment as Pro — one-time $100 credit, then API rates.
• Team Premium: same treatment as Max — 50 per cent of the weekly limit.
• Enterprise: unchanged; direct-contract pricing.

The Pro tier is where the number becomes concrete. A single Fable 5 API session that returns two hundred thousand output tokens costs $10. A $100 credit is ten of those sessions. That is a sentence a support-forum thread will spend the next month arguing over.

The reversal Anthropic won't call a reversal

The delicate part of the Friday thread is that Anthropic had already told the market it was going the other way. When Fable 5 came back on July 1, the plan was to include it in Pro, Max, Team and select Enterprise plans at 50 per cent of the weekly quota — for one week, through July 7, after which it was usage credits for all tiers. The internal thesis was that Fable was too expensive to include in the $20 tier without either a nine-figure inference bill or a queue that would drive Pro subscribers to cancel.

Three weeks later, the calculus flipped. The Max tier gets Fable 5 as a promised feature, not an introductory promotion. The Pro tier gets a credit-then-API off-ramp instead of the promised removal. Anthropic did not use the word reversal in the thread. It used two other words:

"Fable demand has been hard to manage and frustrating for users. We're continuing to invest in more capacity."

And, a few lines down:

"Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages."

Two consecutive sentences that admit demand management and demand prediction are separate problems, each open. That is a candid pair of admissions for a company that spent the June export-control episode explaining that its capacity assumptions had been validated. It is also the kind of language a company deploys when the salespeople who write the SEC-facing forward-looking statements have been on the phone with the pricing team since Tuesday.

Why Monday, not next month

The proximate cause is the calendar. Anthropic's fifty-per-cent bonus to Claude Code weekly rate limits — the temporary uplift that shipped alongside the July 1 Fable 5 redeployment — was scheduled to expire July 20. The choice on July 20 was going to be either announce a permanent structure or roll the expiration through and hope subscribers noticed later. Anthropic chose to announce, and to announce three days in advance, because the second-order effect of not announcing was going to be a support-load spike big enough to move the roadmap.

The distal cause is competitive. In the three weeks between the Fable 5 relaunch and Friday's thread, two facts about the pricing environment changed. First, on June 30 OpenAI shipped GPT-5.6 Sol, priced at roughly a third of Fable 5's per-token rate. Second, Moonshot's Kimi K3 open-weight release now matches or beats Fable 5 on frontend coding benchmarks, at somewhere between zero and a rounding error per token depending on who runs the inference. Alibaba slid Qwen 3.8 onto the same slot twenty-four hours after the Anthropic thread and used the phrase "second only to Fable 5" in the launch press release, which is a marketing choice with two audiences and no interpretive ambiguity.

Anthropic's July 1 pricing was set on the assumption that Fable 5's premium position could carry a $50-per-million-output rate. On July 18, with GPT-5.6 Sol at a third of that and Kimi K3 threatening to be free at the point of use, the tenable pitch had shifted from pay us the premium to stay on the subscription that includes the premium. Half of a smaller cap is the price at which the pitch still works.

The Pro-tier off-ramp is the tell

The Pro tier is the honest sentence in the thread. Anthropic is not lowering its assessment of what Fable 5 costs to serve — the API rate card is unchanged. It is lowering the assumption that Pro subscribers, on average, use enough Fable 5 to justify subsidising them out of the flat $20. The credit-then-API structure lets Anthropic keep the Pro price the same, keep the Fable brand available inside the Pro app, and let self-selection decide who is a $20 customer and who is really a $200 customer that hasn't upgraded yet. It is a segmentation move, executed as a credit.

The $100 figure is calibrated. Ten large-response Fable 5 sessions is enough that a lightly-using developer can not notice the change through August and read the September support-forum post about their credit running out as someone else's problem. It is also enough that a heavily-using developer will hit the wall inside three days, follow the upgrade prompt, and land on Max at a hundred a month. Anthropic just built the funnel it needed to build without adjusting the top-line Pro price by a dollar.

What this means

Three claims, in order of confidence.

1. The Fable-5-in-subscriptions question is settled, and it's settled downward. Anthropic is no longer trying to include the frontier model at every subscription tier. The July 20 structure locks in a two-track world — Max and Team Premium buy included Fable 5, everyone below buys credits. Expect this to be the shape of every future Anthropic frontier launch: the top model available in the top tier at a throttle, and a pay-as-you-go path for the middle tiers. The Loop covered the June programmatic credit pool as a first attempt at that segmentation. July 20 is the second attempt, done more openly, with the same underlying mechanic.
2. GPT-5.6 Sol and Kimi K3 set the ceiling. Anthropic's July 1 pricing was set against a world in which OpenAI's top-tier model was still Mythos and open-weights competition was Qwen 3.5 or DeepSeek V4. Three weeks later the ceiling is a third what it was. Anthropic did not cut the API price on Friday because cutting the API price is the third-order move. The first-order move is to hold the API price and let subscription structure do the work. If GPT-5.6 Sol pricing holds through August, expect the API card to move by the end of Q3.
3. The Pro tier is a churn-risk problem now. Every consumer-AI subscription that has removed a headline feature from the flat rate and replaced it with a credit has, in the twelve to eighteen months following, experienced a measurable churn event on the tier immediately below the one that kept the feature. Anthropic knows this. The $100 credit is a churn hedge, not a subsidy. The question the Pro-tier retention team is now trying to answer is what fraction of Pro subscribers will upgrade to Max at a hundred a month and what fraction will cancel on the same phone call.

What to watch

• Whether Anthropic's next investor communication — the S-1 amendment or an interim ARR-progress statement — discloses the Max-to-Pro subscriber ratio before and after July 20. A visible mix shift toward Max is the metric that makes Friday's thread a win.
• Whether the Fable 5 API price ($10/$50) holds through August. A cut here would concede the GPT-5.6 Sol comparison in a way the subscription move did not.
• Whether Team Premium's ratio to Team Standard widens. If enterprises upgrade their teams for included Fable 5 access, Friday's structure worked; if teams stay on Standard and buy the API on top, Anthropic reproduced the double-payment problem Satya Nadella described as the reason to sell against Claude.
• Whether the July 20 34-per-cent-of-baseline weekly cap — 50 per cent of a limit that dropped 33 per cent — becomes the number a Reddit thread turns into a meme. If it does, the credibility of the "we're continuing to invest in more capacity" line is spent.

Three weeks ago Anthropic told the government the harness could be un-set. On Friday it told its subscribers the feature could be un-set. The two announcements share a grammar. Both are commitments made in the same voice — behavioural, revocable, calendar-anchored. On Monday one of them takes effect. The other one still hasn't been tested.