SemiAnalysis testing found that Claude subscription quotas were lowered by a backend A/B test

Claude Caught Quietly Cutting Quotas! AI Memberships Are Collectively Turning Into Rigged Scales

Ifanr Download the app for exclusive early access October 23, 2026 14:04:48 From Beijing Updated October 23, 2026 15:19:51

Recently, quite a few users have complained that Claude's quota consumption speed has suddenly become extremely abnormal.

Just as everyone was wondering whether they were collectively hallucinating, the well-known semiconductor and AI research firm SemiAnalysis dropped hard evidence.

They posted on X to sound the alarm, confirming that users' feelings were not wrong at all: Anthropic really is quietly running an A/B gray test on Claude's subscription quotas in the background.

On the same day, SemiAnalysis ran an Opus 5.5 Agent workload test on two completely identical $200 Max 20x accounts, and the results revealed a very clear difference.

Test group account B was given a 5-hour window quota of 2.076 billion tokens, 19% higher than the normal account A's 1.74 billion tokens.

But when you look at the more critical weekly total ammunition, account B shrank from the normal 6.579 billion tokens to 5.266 billion tokens, wiping out a full 20%.

Swipe left to see more

Converted into monthly API-equivalent value, normal account A can generate $11,726 worth of compute value per month, equivalent to 59 times the $200 monthly fee.

Meanwhile, the unlucky account B shoved into the gray-test group saw its monthly equivalent value fall to $9,385, with the subsidy multiplier dropping to 47 times the monthly fee. Heavy users are having a full 20% of their quota quietly taken away every month.

No wonder that in the past couple of days, countless Vibe Coding folks who paid for the $100 and $200 plans suddenly found that the Opus 5.5 in their hands had completely changed flavor.

People used to joke that ChatGPT's catchphrase was "You're right." Now, for Claude Code, its catchphrase has directly become "You've reached your usage limit."

Another X user reported that last week they could barely feel any limits, but today, before they had even run many heavy tasks, 23% of their quota was already gone.

Another user named Priya was even more helpless: their daily work had just gotten started, and the 90% progress bar was already at the bottom.

What's even more interesting is the official damage-control posture right away.

Facing questions from across the internet, Claude Code engineer Lydia Hallie came out to reassure everyone, explicitly stating that there had been no adjustment to quotas and suggesting that users who felt consumption was fast should run the feedback command and DM her their IDs for investigation.

Right after that, she offered a set of explanations shifting the blame, asking whether users often left in the middle of long conversations for more than 1 hour, causing the context cache to expire, or whether they had opened too many sub-agents in a session.

Claude Code's core developer Boris Cherny has also just stated that the team is urgently investigating the cause and will give everyone an explanation soon.

The most unbearable part of the whole affair is that just half a month ago, SemiAnalysis had just given Anthropic a very valuable good-guy card.

At the time, in order to dig out the real dilution behind each company's plans, the analyst wrote a test program that repeatedly stuffed the text of the classic novel War and Peace into major models, specifically watching that blind-box progress bar that only shows 0% to 100% but never specifies the exact number of tokens.

After a round of extreme stress testing and calculation, among plans in the same $200 tier, the API-equivalent value provided by a Claude subscription using Opus 5.5 was about 5 times that of OpenAI's same-tier plan using GPT-6.1 Sol.

The direct trigger for the gap widening to a runaway 5x lead was that neighboring OpenAI struck earlier and with even less restraint.

OpenAI had previously cut the token quotas for each model tier in its $200 Pro plan in half on the spot, and even erased the relative-multiple promotional language from its official pricing page.

After cutting the $200 plan, OpenAI turned around and rolled out a new $500-per-month tier.

But actual testing showed that for the extra $300, the top-tier Astra model's quota was only about 21% higher than the pre-adjustment $200 plan. The real gimmick became the ultra-fast output mode of 300 tokens per second.

In short, the hundreds of extra dollars users pay do not buy much additional quota; a large part of it is simply paying for output speed. Now, the seemingly upright Anthropic has ultimately taken the same approach.

The reason may be that, in the eyes of vendors such as OpenAI and Anthropic, the heavily subsidized monthly subscription system has become the biggest drag on profit margins.

According to SemiAnalysis's calculations, the subscription business contributed only about 10% of Anthropic's total revenue but consumed more than 40% of its inference compute.

But does the large-model business really have to sustain gross margins only by continuously shrinking user quotas? Turning our gaze back home, look at the engineering answer just delivered by domestic model vendors—it is a completely different picture.

Just in the past couple of days, DeepSeek published its technical practice for large-scale inference deployment of V4.1 Flash and officially open-sourced its cloud service inference framework DeepInfer. Using common 8-GPU machines for prefill and 64-to-128-GPU supernodes for decoding, a single DeepSeek cluster achieved a daily throughput of 155 trillion tokens.

There are two very hardcore data points in this report.

The first is that, at the system design level, DeepSeek supports retaining KVCache context caches for more than 72 hours, solving the cost problem of long-cycle context reuse with solid engineering optimization.

The second directly lays bare the money-making ability of this business: even calculated at off-peak low prices, the overall revenue of DeepSeek's inference system reached $38 per GPU per hour.

Do the simplest math: one GPU can generate more than $900 in revenue in 24 hours a day, fully proving that as long as the underlying inference architecture and scheduling optimization are done well, the AI model business can absolutely achieve extremely high gross margins.

In the mobile internet era, the marginal cost of software and content services is almost zero. After users subscribe to a monthly membership, no matter how frequently they use it, the extra bandwidth cost the platform pays is negligible.

And what makes users most uncomfortable may not be the price increase itself, but the lack of basic decency and transparency in the whole process.

Vendors do not publish specific consumption standards, only giving a vague percentage progress bar, either secretly degrading intelligence or quietly running A/B tests in the background to lower quotas.

Users pay real money in subscription fees on time every month, yet the service standard they buy may be tampered with at any time without their knowledge. This kind of trust overdraft, treating users as fools to be played, is probably harder to make up for than compute consumption.

"Special statement: The above work content (including videos, pictures, and audio if any) is uploaded and published by users of Dafeng Hao, a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."