Why is the "AI service subscription model" doomed to fade away?

Title: "Why Subscription-Based AI Services Are Doomed to Disappear"
Author: Zhang Yongyi, Geek Park

Author: Rhythm BlockBeats

Source:

Reprinted from Mars Finance

Subscription models will be drained and should be cherished.

On June 9, Anthropic released its strongest public model to date, Claude Fable 5.
As usual, this should be a festival for paying users—your monthly payment finally grants you access to the flagship model at the earliest opportunity.

But one line in the announcement immediately sparked huge controversy after release:
After June 22, Fable 5 will be removed from all subscription plans, and continued use will require purchasing usage points separately.

In other words, even if you buy a membership, the flagship model is only available for 14 days.

A model that comes with a "kick-out order" on the day of release—this is the first time in the large model industry.

Many people see it as a mistake or arrogance on Anthropic’s part. My view is the opposite: this is not a mistake, but a warning.

The AI subscription model is heading toward an inevitable demise—not because any company is greedy, but because the very premise it relies on is being dismantled by AI itself.

01 The 14-Day Countdown for the Flagship Model

First, let’s clarify the facts. According to Anthropic’s official plan (June 9, 2026),
Fable 5 is included free of charge in the Pro, Max, Team, and seat-based enterprise plans from the release date until June 22;
Starting June 23, it will be removed from these plans, and every subsequent token will be deducted from pre-paid usage points, at rates identical to the API.

This rate is not cheap: $10 per million input tokens, $50 per million output tokens, exactly double that of the previous flagship Opus 4.8.
More subtly, even during the free window, Fable 5 is calculated at roughly double the weight within subscription quotas—doing the same work consumes twice the quota compared to Opus.

User reactions are predictable. Someone on Hacker News directly said that this "pay first, then charge" approach is unsettling, suspecting Anthropic aims to push subscription users toward pay-as-you-go billing;
Developers have tested that on the $100/month Max plan, a single agent programming session consumes nearly $100 worth of tokens.

And this is not just Anthropic’s move. Over the past eight weeks, the entire industry has been doing the same:
On April 2, OpenAI switched Codex from message-based billing to token-based billing aligned with the API, then extended this to all existing enterprise customers.

GitHub froze new sign-ups for Copilot Personal on April 20, and a week later announced a full switch to AI Credits billing, completed by June 1—$10 monthly fee plus $10 credits.

Anthropic’s own actions have been the most intense:
From April 4, banning third-party agent frameworks like OpenClaw from consuming subscription quotas, shifting such usage to pay-as-you-go;
On April 21, the Pro plan’s Claude Code section on the pricing page was quietly marked with a red cross, then retracted within 24 hours after community backlash, with the official explanation being "a small test for about 2% of new users";
On May 14, it was officially announced that from June 15, Agent SDK and interface-less calls would be removed from the subscription pool and replaced with independent points billed at API rates.

Three companies, eight weeks, the same direction—this is no coincidence, but the industry collectively answering the same math problem.

What does that math problem look like?

02 Pricing Has Never Been About Computing Power

Research firm SemiAnalysis recently laid out this math problem.
They bought each subscription tier from Anthropic and OpenAI, ran long-term programming tasks until weekly limits were exhausted, then converted usage into dollar value based on API prices: how much is this usage worth?

The industry’s previous understanding was that a $200/month plan could support about $2,000 worth of tokens.
Actual tests far exceeded this:
Claude Pro at $20 supports about $400;
Max 20x at $200 supports about $8,000.

OpenAI’s figures are even more exaggerated—
ChatGPT Plus at $20 can generate about $700 worth of tokens;
Pro 20x at $200 supports about $14,000.

Two fair points must be made upfront:
These are the upper limits for "full quota utilization," not the typical daily usage for ordinary users;
API prices include profit margins, and the converted numbers do not reflect actual computational costs.

But pricing must set an upper limit—an insurance company cannot assume no claims.

Subsidies themselves are not deadly.
Media subsidies, ride-hailing subsidies, burning money for growth—these are the traditional tricks of the internet industry.
The real danger lies in the fundamental difference between AI subscription models and these subsidies.

Netflix dares to sell monthly packages because two things are true:
The marginal cost of adding another movie approaches zero, and a person only has 24 hours a day to watch.
Spotify is similar.
The implicit premise of monthly subscriptions is that consumption is limited by human physiological constraints—its true pricing is never content, but human time.

In the chatbot era, AI somewhat fits this premise.
A person can chat, but their daily keyboard input is limited;
Idle quotas from light users are enough to cover the overuse by heavy users.

Then, Agent arrived.

What does an agent task look like?
It reads 20 files, plans, modifies code, runs tests, reads errors, and iterates—each round consumes 5 to 30 times more tokens than normal conversations.
More critically, it doesn’t require your presence.

I personally experienced this: recently, I asked an agent to organize flight data for two airports.
I went for a shower, came back, and the task was done, with quotas nearly exhausted.
While you sleep, the electric meter keeps spinning.

Agents don’t cancel the price ceiling—they cancel the consumption ceiling.
And the entire evolution of the AI industry—longer tasks, more autonomy, multiple parallel instances—are all rushing toward the same endpoint:

Removing humans from the consumption process altogether.

GitHub plainly states in its announcement that "agent usage is becoming the default."
This means that the scenarios where subscription still makes sense—people sitting in front of screens chatting sentence by sentence—are shrinking, and their share in AI’s value landscape will only get smaller.

At this point, some might ask:
If subsidies are so deep, why not just raise prices?

It’s been tried, and the result is worse.
Looking back at SemiAnalysis’s table, there’s an odd detail: the more expensive tiers have higher subsidy multiples.

For Claude, the $20 tier’s multiple is 20x, the $200 tier’s is 40x;
For OpenAI, it’s from 35x to 70x.
Half of this is by design—higher tiers amplify quotas proportionally, effectively giving discounts to big clients;
The other half is user behavior—those willing to spend $200 on a 20x plan are chasing full utilization, while light users wouldn’t appear in this tier at all.

This is called adverse selection in insurance—when a policy’s pricing attracts only the highest-risk policyholders, it becomes unprofitable.
Any fixed price will precisely select users who exceed their quotas—this is not a management problem, but a structural one.
Adjusting prices only makes the sieve finer.

Throughout 2025, the industry tried every patch.
In January, Sam Altman admitted on X that the $200/month ChatGPT Pro was losing money because usage far exceeded expectations—price hikes failed.
Mid-year, Cursor switched from request-based to compute-based billing, triggering mass cancellations, and the CEO publicly apologized—changing rules midstream failed;
In summer, Anthropic added weekly limits to Claude Code, citing users running agents around the clock, with individual consumption reaching tens of thousands of dollars—rate limiting only provoked anger.

After all patches failed, the industry faced a collective showdown over the past eight weeks.
OpenAI’s ChatGPT lead Nick Turley clarified on BG2 podcast:
"In today’s era, offering unlimited plans might be like offering unlimited electricity."

03 The Shell Is Still There, But the Core Is Dead

Of course, there’s a powerful counterargument:
Subscription models are still thriving.
ChatGPT Plus remains $20/month, Claude Pro is still on sale, GitHub’s code completion still offers monthly plans.
Is the idea of extinction just alarmist?

This counterargument deserves serious consideration because the phenomena it describes are real.
But it misses what is truly dying.

The soul of subscription models has never been the "monthly deduction" format, but the promise of "fixed price, worry-free use"—you don’t have to calculate each usage cost, and that was the whole reason it beat pay-per-use.

What’s happening now is: the deduction cycle remains, but the promise is being stripped away.

In GitHub Pro’s $10 monthly fee, what’s loaded are $10 worth of points, used up as you go—this is not a subscription, but a prepaid recharge card disguised as one.
Anthropic’s points are deducted at API rates, OpenAI’s points support auto-recharge.
Subscription models are not disappearing—they are being hollowed out: the shell remains, but the core is dead.

The only real enclave left: pure chat.
It can still be monthly because it’s the last scenario where human time still limits consumption.
But that moat can’t hold—every penny of R&D in this industry is pushing AI from "you ask, it answers" toward "it proactively helps you complete tasks."

Chat subscriptions won’t be killed—they will be marginalized:
Staying put, watching real value and revenue slowly migrate into pay-as-you-go territory.

Another hard-to-ignore coincidence: according to TechCrunch (June 2026),
at the launch of Fable 5, Anthropic is reportedly preparing for an IPO alongside OpenAI.
Over the past three years, subsidies have been funded by venture capital; public market investors won’t accept a profit-and-loss statement that shows "more heavy users, more losses."
The timetable for capital retreat determines that the showdown won’t be indefinitely delayed.

This means different things for different people.
For companies, AI spending will need to be managed like cloud expenses—
According to The Information, Uber’s CTO told internal memos that the company burned through its entire 2026 AI budget in four months, making budgeting, monitoring, and routing models by task a must-have skill for every team.

For individual users, the past was light users subsidizing heavy users; now, everyone pays for their own electric meter.

Honestly, this might not be all bad.
Price signals returning mean "Is this task worth running AI for?" becomes a real question—
And when an industry begins to seriously answer that question, it’s often the start of moving away from money-burning narratives toward normal business.

At this point, I want to add: before the electric meter was installed, the current subscription model might be the industry’s most generous moment—use it, cherish it.

The logic is hidden in SemiAnalysis’s table.
From the user’s perspective, it’s not a death sentence but a list of ongoing benefits:
Pay $200/month, and the platform burns up to $14,000 worth of compute for you.

Such a subsidy level hasn’t appeared since the ride-hailing and food delivery wars—
And we remember how those wars ended: subsidies disappeared, prices never returned.

So, the heavy tasks should be run now.
For example, Fable 5’s subscription window only lasts until June 22.
Rather than waiting for the points era to come and carefully planning, it’s better to schedule those long tasks you’ve always wanted to run but found too expensive.
This isn’t about exploiting the system—just being a clear-eyed beneficiary of a pricing mistake that’s bound to be corrected.

Turley’s metaphor might be deeper than he intended:
The true sign that electricity has become infrastructure isn’t that it reaches every home, but that every home has an electric meter—
From that moment, no one discusses "monthly electricity," only electricity prices.

Subscription models won’t have a funeral.
They will quietly turn into a line item called "Entry Fee" on your billing statement on some quiet billing day.

Until then—cherish, and use it well.