The subscription model will be hollowed out and cherished.
On June 9th, Anthropic released its most powerful public model to date, Claude Fable 5. As usual, this should have been a festival for paying users—your monthly payment has finally earned you the privilege of being the first to experience the flagship model.
However, a line in the announcement immediately sparked controversy upon its release: after June 22nd, Fable 5 will be removed from all subscription plans, and continued usage will require separately purchased usage credits.
In other words, even if you are a member, the flagship model will only be available to you for 14 days.
A model coming with a "pass" on the day of its release is unprecedented in the large model industry.
Many view this as a mistake or an act of arrogance on Anthropic's part. My view is the opposite: this is not a mistake but a warning.
The AI subscription model is heading towards an inevitable demise—not because any company is greedy, but because the very premise on which the subscription model is based is being dismantled by AI itself.
01 Countdown to a 14-Day Flagship Model
Let's get the facts straight. As per Anthropic's official arrangement (June 9, 2026), Fable 5 is freely included in the Pro, Max, Team, and seat-based Enterprise editions from the release date until June 22nd; starting from June 23rd, it will be removed from these plans, and each subsequent token will be deducted from pre-purchased usage credits at a rate consistent with the API.
This rate is not cheap: $10 per million input tokens, $50 per million output tokens, precisely twice the rate of the previous flagship Opus 4.8. More subtly, even during the free window, Fable 5 is calculated in the subscription quota at roughly twice the rate—doing the same work burns through the quota at double the speed of Opus.
User reactions are predictable. Someone on Hacker News bluntly stated that this "give first, take later" operation is unsettling, suspecting that Anthropic aims to use this opportunity to push subscription users towards usage-based billing. Some developers have tested and found that on the $100 per month Max plan, a single agent programming session consumed nearly $100 worth of tokens.
Moreover, this is not just Anthropic's move. Over the past eight weeks, the entire industry has been doing the same thing: on April 2, OpenAI switched Codex from message-based billing to API-aligned token-based billing, then extended it to all existing enterprise customers.
GitHub froze registrations for Copilot Personal on April 20, announced a week later that they were moving entirely to AI Credits billing, and completed the transition on June 1—$10 per month for the Pro tier, which comes with $10 in credits.
Anthropic itself had the most intense actions: starting April 4, third-party agent frameworks like OpenClaw were prohibited from consuming subscription quotas, shifting such usage to pay-as-you-go; on April 21, the Claude Code field under the Pro plan on the pricing page quietly turned into a red 'X,' which was retracted within 24 hours after community uproar, with the official explanation being a "small test for about 2% of new sign-ups"; on May 14, it was officially announced that starting June 15, the Agent SDK and headless calls would be removed from the subscription pool and converted to a separate point-linked API rate
Three companies, eight weeks, one direction—this is not a coincidence, it is the entire industry handing in the same answer to the same mathematical problem.
What does that mathematical problem look like?
02 Pricing Has Never Been About Compute
Research firm SemiAnalysis recently put this math problem on the table. They bought each tier of subscription from Anthropic and OpenAI, ran long programming tasks until they depleted the weekly limit, and then converted it into API list price: how much are these usage levels worth.
The prevailing industry belief was that a $200 monthly package could at most generate around $2000 worth of tokens. The actual test results far exceeded this: a $20 Claude Pro, with a limit of about $400; a $200 Max 20x, around $8000.
OpenAI takes it to the next level—$20 ChatGPT Plus can generate about $700, $200 Pro 20x, around $14,000.
Two fair points to note upfront: this is the maximum value of "running to the limit," not the everyday waterline for regular users; API pricing includes a margin, and the equivalent figures do not represent the actual cost of computation power.
But pricing must be anchored to the upper limit—insurance companies can't assume that nobody will make a claim.
The subsidy itself is not lethal. Streaming services have been subsidized, ride-hailing apps have been subsidized, burning cash for growth is an ancestral skill of the internet. What is truly fatal is the AI subscription model and the fundamental difference between them.
Netflix dares to sell monthly subscriptions based on two things: the marginal cost of adding one more movie is close to zero, and a person can only watch a maximum of 24 hours a day. The same goes for Spotify. The implicit premise of the subscription model is that consumption is locked by the physiological limit of a person—its real pricing has never been about content but about people's time.
AI in the era of chatbots barely fits this premise. However much a person can chat, there is a limited amount of keyboard typing in a day; the surplus allowance of light users is enough to cover the overconsumption of heavy users.
Then, the Agent arrives.
What does an agent task look like? It reads 20 files, does planning, changes code, runs tests, reads error messages, then iterates—after one round, token consumption is 5 to 30 times that of a normal conversation. What's more lethal is that it doesn't require your presence.
I have had a personal experience of this: not long ago, I asked an agent to organize the flight data of two airports. I took a shower, and when I came back, the task was completed, and the quota was depleted. While you were sleeping, the meter was running.
What the agent canceled was not the price cap but the consumption cap. And all the evolutionary directions of the AI industry—longer tasks, more autonomy, multiple instances in parallel—are sprinting toward the same endpoint:
To completely remove humans from the consumption process.
GitHub explicitly stated in its announcement that the use of agents is "becoming the default." This means that even the scenarios where subscription models can barely survive, where people chat one by one in front of the screen, will account for a smaller and smaller proportion of AI's value proposition.
At this point, someone might ask: "If subsidies are too deep, why not just raise the price?"
We tried that, and it led to an even worse outcome. Looking back at SemiAnalysis' chart, there is an abnormal detail: the higher the tier, the higher the subsidy multiplier.
On Claude's side, the $20 tier has a multiplier of 20x, the $200 tier has 40x; on OpenAI's side, it increased from 35x to 70x. Half of this is due to pricing design—magnifying the quota for higher tiers is like giving a discount to large customers; the other half is due to user behavior—those who are willing to spend $200 on a 20x package are aiming to max it out, and light users will never appear in this tier.
This has a name in the insurance industry: adverse selection. When the pricing of a policy attracts only the highest-risk policyholders, that policy has no actuarial viability. Any fixed price will accurately filter out the group of users who exceed it—this is not a business issue but a structural problem, and adjusting the price will only make the filter finer.
Throughout the whole year of 2025, the industry tested every patch. In January, Sam Altman admitted on X that ChatGPT Pro, priced at $200 per month, was running at a loss because usage far exceeded expectations—the price increase tier failed.
Mid-year, Cursor changed from request-based billing to compute-based billing, causing a large number of unsubscribes. The CEO publicly apologized—an unsuccessful mid-rule change; in the summer, Anthropic introduced a weekly limit to Claude Code because some users were running the agent 24/7, consuming computing resources worth thousands of dollars per person—the flow restriction only brought anger.
After all patches were rendered ineffective, we saw this eight-week-long showdown this year. Nick Turley, the head of ChatGPT at OpenAI, made it clear on the BG2 podcast: "In the current era, offering an unlimited plan may be akin to offering an unlimited electricity plan."
The Shell Remains, but the Core is Dead
Of course, there is a seemingly strong counterargument: isn't the subscription model doing just fine? ChatGPT Plus still costs $20 a month, Claude Pro is still on sale, and GitHub's code completion even retained its monthly package. Is the talk of demise an exaggeration?
This counterargument deserves serious consideration because it describes a real phenomenon. However, it misinterprets what has actually died.
The essence of the subscription model has never been the "monthly deduction" format; it has always been the promise of a "fixed price, worry-free usage" — you don't have to calculate the cost of each use, which was precisely why it triumphed over pay-per-use back in the day.
What is happening now is that: The billing cycle remains, but the promise has been stripped away.
Within GitHub Pro's $10 monthly fee, you are essentially purchasing $10 credits, which expire after use — this is not a subscription; it's a prepaid credit card disguised in a subscription's clothing. Anthropic's credits are deducted per API rate, and OpenAI's credits support auto-recharge. The subscription model will not be canceled; it will be hollowed out: the shell remains, but the core is dead.
There remains one true enclave: pure chat. The reason it can still offer a monthly package is that it is the last scene in AI where consumption is still time-locked by humans. However, no moat can protect an enclave — every penny of R&D in this industry is pushing AI from "you ask, it answers" to "it actively helps you complete tasks".
Chat subscriptions will not be killed off; they will be marginalized: staying in place, watching as true value and real revenue slowly move into the pay-as-you-go world.
There was another coincidental point in time that is hard to ignore: according to a TechCrunch report (June 2026), around the release of Fable 5, Anthropic is preparing to go public alongside OpenAI. Over the past three years, subsidies have been footed by venture capital; public market investors will not stomach a P&L that loses money for each heavy user added. The timeline for capital exit dictates that the reckoning will not be indefinitely delayed.
This means different things to different people. For enterprises, AI spending will now need to be managed like cloud spend — according to The Information, Uber's CTO said in an internal memo that the company burned through its entire 2026 AI budget in four months, budgeting, monitoring, and task-routing models will become mandatory coursework for every team.
For individual users, where in the past light users subsidized heavy users, now, everyone pays for their own meter.
Frankly speaking, this is not necessarily all a bad thing. After the price signal regression, the question of "is this task worth running with AI" has become a real issue for the first time — and when an industry seriously starts answering this question, it is often the beginning of its departure from a narrative of burning money and a move towards normal business.
At this point, I'd like to interject: Before even the meter is installed, the current subscription model may be the industry's most generous moment for users — use it and cherish it.
The logic is hidden in SemiAnalysis's table. Reading it from the perspective of a user, it is not a death sentence at all, but a welfare list that is still in effect: you pay $200 per month, and the platform covers up to $14,000 of computing power for you.
This level of subsidy was last seen during the ride-hailing and food delivery wars — and we all remember the outcomes of those two wars; once subsidies exit, prices never go back.
So the heavy lifting that needs to be done, do it now. For example, the window for Fable 5 to remain in the subscription is only until June 22, so instead of waiting for the era of points to arrive and then carefully calculating, it's better to schedule those long tasks that you've always wanted to run but found too expensive. This is not taking advantage — just being a clear beneficiary in a pricing error that is bound to be corrected.
Turley's analogy may have a deeper meaning than he intended. The true sign of electricity becoming infrastructure was not when it reached every household, but when every household had an electric meter installed. From that moment on, no one discussed whether electricity should be metered; instead, they talked about the price of electricity.
A subscription model will not have a eulogy. It will only, on a quiet billing day, become a small line item in your expenditure details called an "entry fee."
Until then—use it and cherish it.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
