There is a particular kind of corporate comedy that only writes itself, and Anthropic just delivered the third act.
On July 12, for the second time in a week, the company pushed back the deadline for pulling Claude Fable 5 out of its paid subscriptions. The new cutoff is July 19, 11:59:59 PM Pacific. That is the third date shift in five weeks — July 7, then July 12, then July 19 — each one framed as a generous act of mercy, each one dripping with the same energy as a furniture store's "FINAL DAYS! (again)" banner $TRAE_REF.
The official reason is "capacity." The actual theater is scarcity marketing. Anthropic wants you to feel that Fable 5 is precious, fleeting, rationed — a luxury you must hurry to consume before the gates close, and afterward pay $10/$50 per million tokens to keep touching. Usage credits. Twice Opus 4.8's rate. Anthropic's highest published price for a generally available model $TRAE_REF.
And here is the punchline that landed while they were busy printing the banners.
Artificial Analysis has scored Moonshot AI's Kimi K3 at 57 on the Intelligence Index. That puts an open-weights model — weights scheduled to ship free on July 27 — ahead of Anthropic's shipping flagship Claude Opus 4.8 at 56, one point behind GPT-5.6 Sol at 59, and three behind Fable 5 at 60 $TRAE_REF.
Read the ordering again. The model Anthropic ships as its best — Opus 4.8, the one sitting inside every Pro and Max plan as the default workhorse — now loses, on a third-party benchmark, to a model a high schooler can download and host. The only Anthropic model still ahead of K3 is Fable 5. The one they are about to yank from your subscription.
Now do the July 19 math.
When the Fable 5 window closes, the best model remaining inside a $200 Max plan is Opus 4.8. Opus 4.8 scores 56. Kimi K3 scores 57. K3 is open weights. K3 costs an average of $0.94 per task on the Intelligence Index, against Opus 4.8's $1.80 — roughly half the cost, and it scores higher $TRAE_REF. So the plan Anthropic sells as "frontier access for $200 a month" will, on Sunday morning, offer a best in-house model that is beaten by a free model running at half the per-task cost.
This is not a pricing problem. This is a narrative collapsing on itself in public.
Let's walk through the sequence, because the comedy only works if you watch every beat.
June 9. Fable 5 launches. Anthropic prices it at $10/$50 per million tokens — double Opus 4.8 — and quietly bundles a promotional slice into Pro, Max, and Team plans. "Try our most capable model, included for a limited time." $TRAE_REF
June 12. A Commerce Department export-control order takes Fable 5 and Mythos 5 offline for 19 days. Anthropic does not lower the price. The model is simply gone. Subscribers, who were paying for "frontier access," get nothing in return except an apology and the existing models they were already paying for.
July 1. Fable 5 returns after the controls lift. But the terms have changed. Subscribers can use it for "up to 50% of their weekly usage limits." Past that, pay usage credits. The clock starts. Seven days.
July 7. First deadline arrives. Anthropic extends to July 12. "We need more time," is the framing, dressed up in "as soon as capacity allows." $TRAE_REF
July 9. OpenAI ships GPT-5.6 Sol to general availability. Artificial Analysis scores it 59, one point behind Fable 5, at roughly a third of the cost per task. $TRAE_REF
Hours later. Anthropic resets all five-hour and weekly rate limits for every user. No explanation connecting the two. OpenAI's Tibo Sottiaux posts three words on X: "I smell fear." $TRAE_REF Anthropic does not reply.
July 12. Second extension. July 19. Same "capacity" language. Same scarcity framing. Same implicit message: hurry, the door is closing, buy credits.
This is what fear marketing looks like when it forgets to check whether the thing it's gatekeeping is still the best in the room. Anthropic spent five weeks manufacturing urgency around a model that, by the final extension, had an open-weights competitor within three points — and the competitor was overtaking the model that would replace Fable 5 in the plan they were selling.
Here is where it goes from bad marketing to genuinely funny.
The whole point of pulling Fable 5 from subscriptions was to move paying users onto usage credits. $10 per million input. $50 per million output. The justification is that frontier intelligence costs money to serve, and Anthropic cannot afford to give it away inside a flat-rate plan forever. Fair enough, as a business argument, in a world where Fable 5 is unambiguously the best model money can buy.
But that world expired on July 16, when Kimi K3 landed at 57.
After July 19, the value proposition of a $200 Max subscription is no longer "access to the frontier." It is "access to Opus 4.8, which is one point behind a free model." The thing Anthropic is gatekeeping — Fable 5 — is no longer the frontier anyway; GPT-5.6 Sol is one point behind at a third of the cost, and the open model is three points behind at half the cost. The moat Anthropic is selling access to has, in the span of a week, become a model that is third place on the index and first place on price-to-performance.
The fear marketing assumed subscribers would pay anything to keep the "best model." It did not account for the possibility that the best model in the plan would no longer be the best model, period. You cannot charge frontier prices for sub-open-source intelligence. And yet, here is the rate card, ready to go on Sunday morning: $10/$50 per million tokens, for a model that is three points ahead of a model you can download.
The three extensions are the real punchline, because each one was framed as generosity while functioning as panic.
When you extend a "limited time offer" once, you are managing capacity. When you extend it twice in a week, you are managing perception. When you extend it a third time five weeks out, you are a department store with a "GOING OUT OF BUSINESS SALE (we promise this time)" banner, and the customers have started to notice the banner has been there since June.
The stated reason — "as soon as capacity allows" — is the kind of non-statement that only works if nobody is looking at the scoreboard. The scoreboard says: a Chinese lab with an open-weights model just matched your shipping flagship's intelligence at half the per-task cost. Capacity is not your problem. The problem is that the thing you are rationing is no longer scarce, and the thing you are about to leave in the subscription is no longer frontier.
The OpenAI "I smell fear" tweet was cruel because it was accurate, and Anthropic's silence confirmed it. You do not reset every user's rate limits hours after a competitor ships a cheaper near-parity model unless the reset is, at minimum, a competitive twitch. Then you extend the Fable 5 window again three days later. These are not the moves of a company with a capacity problem. These are the moves of a company trying to keep a story alive past the point where the story is true.
On July 20, the Monday after the cutoff, a $200 Max subscriber opens Claude and finds that Fable 5 is gone. The best model in the plan is Opus 4.8. Opus 4.8 scores 56. A model that anyone can download, host, or call for $0.94 a task scores 57. The subscriber is now paying $200 a month for the privilege of losing to free.
The honest version of the post-July-19 pitch is: "Subscribe to Claude Max, and you too can be beaten by a model a teenager in Shenzhen is running on rented GPUs." That is not a pitch that sells. It is a pitch that gets screenshotted and put in a viral thread next to Kimi K3's $0.30/MTok cache-hit price.
Which is, as mentioned, roughly half the per-task cost of Opus 4.8 — and now ahead of it on the index. So the "almost half the cost" framing in the original premise undersells it. The open model is cheaper, smarter, and downloadable. The closed model is the one you get to keep paying $200 a month to access, as a reward for your loyalty, while the frontier moves to a place your plan no longer reaches.
Fear marketing works when the thing you are making scarce is also the thing nobody else can make. That is no longer the condition of the frontier-model market. The frontier is now a gradient, and an open-weights model from Beijing is sitting one point behind the most expensive closed model on Earth, at half the cost, with the weights dropping free in ten days.
Anthropic's play was to ration Fable 5 like it was the last frontier in town. The town just got a new neighbor, the neighbor is giving the stuff away, and the neighbor's free version just beat the model Anthropic is leaving in your $200 plan. The extension to July 19 was supposed to be a mercy. It turned out to be a countdown to the moment the subscription's value proposition quietly died.
You cannot charge frontier prices for sub-open-source intelligence. The fear marketing was bad last week. It is impossible now. And the funniest part is that Anthropic, with three extensions and a rate-limit reset, is the one who told everyone to look at the clock — right as the clock was running out on the story they were selling.
Three extensions. Five weeks. One open model. The banner still says "FINAL DAYS." The scoreboard says something else entirely.