Why does a $1.50 ChatGPT Plus account with a three-hour warranty still sell?
Why does a $1.50 Plus account with a three-hour warranty still attract buyers? I follow how sellers transfer risk, buyers rationalize the bargain, and account pools keep the market running.
I started paying attention to this story because of a product priced at 39,999 VND, roughly $1.50. It was called ChatGPT Plus, looked like Plus in the screenshots, and apparently worked with Codex. The only detail that sounded almost like a confession was the seller's warranty of exactly three hours.
Three hours. The official subscription lasts a month, while the seller's promise might not survive an afternoon. What interested me was not only where the account came from or which router ran behind it. The more uncomfortable question was why a seller who knew the product might die so quickly would keep selling it, and why a buyer who saw that condition would still pay.
The further I followed the shops, the less the answer looked like one evil side and one foolish side. This market has converted risk into part of the price. The seller keeps the payment and limits responsibility to three hours. The buyer trades a low price for immediate access, then accepts the uncertainty over how long the account will last, whether the buyer’s data stays private, and whether the account will be suspended. Account pools and routers are the machinery that lets this exchange happen more often.
This article therefore asks more than how a $1.50 ChatGPT Plus account is produced. I want to understand why a murky business can look ordinary enough to operate in public, retain customers, and keep resellers ordering the next batch. The answer sits on both sides of the buy button.
A three-hour warranty does not stop the sale. It makes the sale possible
The same ChatGPT Plus label across the shops I looked at ranges from a few tens of thousands to more than one hundred thousand VND. The cheapest products usually come with a few hours of warranty or none at all. More expensive ones are described as durable, tied to a different mailbox or payment route, or more stable with Codex. Some shops deliver an email address, password, and 2FA code. Some activate Plus on the customer's own account. Others deliver no account at all, only a key and base URL for calling a model.
If all of these were truly the same first-party Plus subscription, the delivery methods would not differ this much. In the market, "Plus" has been stretched into a very broad label. It can mean a real subscription with an uncertain lifetime, a trial, a promotion, a workspace seat, a credential that has passed through other hands, leftover quota, or permission to send requests through a pool. Buyers see the same logo and think they are comparing prices. In reality, they are comparing different products carrying the same name.
The seller is therefore pricing more than "one month." They are pricing the chance that an account survives the next sweep, the number of days they will replace it, whether it clears phone verification, how much Codex quota remains, and how much control they have over the upstream source. It is like buying food with no expiry date on the package. A three-hour warranty is the expiry date written in small print.
Oddly enough, three hours can still create a little confidence. It is more concrete than "no warranty" and gives the buyer time to log in, see the Plus badge, test Codex, and believe the shop delivered. If the account survives for a week, the buyer feels lucky. If it dies tomorrow, the seller points back to the terms. An extremely short promise becomes both sales proof and an escape from responsibility.
A three-hour warranty is not a silly defect in the business model. It is the model's unwritten contract. The seller offers a chance to use the product now, and the buyer accepts almost every consequence that arrives later.
Sellers profit because the loss already belongs to the customer
A product with high attrition can remain profitable if the seller does not absorb every failure. Supply can be acquired cheaply, graded, sold quickly, or placed in a pool to extract each remaining unit of quota. Shorter warranties mean fewer replacement accounts. When an account dies in hour four, the event is no longer part of the shop's cost. It becomes the customer's bad luck.
This is the part I find hardest to excuse. Sellers usually know more than buyers about the source, survival rate, latest enforcement sweep, number of concurrent users, and whether an upstream supplier will provide replacements. The customer sees a logo, a price, and a few testimonials. That information gap is not incidental to the transaction. It is where much of the margin is created. The person who knows the product is fragile sells confidence to someone with no practical way to inspect it.
The model also benefits from volume. A sales bot does not get tired, stock can refill automatically, one reseller can pass on another reseller's supply, and an account inside a pool may serve many requests before it dies. Even a small margin per order is attractive when the operation is automated and customers keep returning. A dead account can even generate the next purchase if the buyer concludes that the previous failure was only bad luck.
I consider profits derived from hidden sourcing, access the seller does not own, or deliberate customer confusion illicit, without assuming that every seller obtains accounts through one method. A retailer at the bottom of the chain may genuinely know little about the upstream source. Deliberately refusing to ask does not make the product ethical. When someone repeatedly takes money for access they know violates service terms, lacks stability, and has no explainable provenance, ignorance has become a business decision.
Why do buyers keep lining up?
The first answer is practical. The official price can be substantial for students, early-career workers, or anyone who needs Codex for one assignment or a short project. Forty thousand VND cannot buy certainty, but it may buy one productive evening as a deadline approaches. The buyer does not necessarily think they are purchasing a complete month of Plus. They may need the account to stay alive only long enough to solve today's problem.
The low price also makes failure easy to absorb. Losing a full subscription price encourages questions about provenance, terms, and privacy. Losing $1.50 can feel closer to buying a disappointing snack. The small ticket does not remove risk. It makes investigating the risk feel more expensive than the purchase. If the account dies, buying another one appears faster than demanding an answer.
A three-hour warranty fits that psychology almost perfectly. The customer gets enough time to verify that the product is alive at delivery, while the future becomes a lottery prize. An account surviving three days feels like a major win even though it falls far short of the advertised month. Once the reference point moves from the official price to 39,999 VND, the buyer stops asking whether they received the full product. They ask how long it must work before they break even.
Group testimonials and screenshots of live accounts add social proof. A buyer who has just logged in has a reason to post a thank-you screenshot. Someone whose account dies after the warranty often goes quiet, calls it bad luck, or buys from another batch. Success remains on the storefront while failure disappears into support messages. A new customer sees a satisfied crowd without seeing everyone who left.
Not every buyer is deceived. Some understand the gamble, know the account may die, and still decide that a few hours of Codex are worth $1.50. Others reassure themselves that OpenAI is a large company, the seller is responsible for sourcing, and one small purchase cannot hurt anyone. Those arguments turn an ethically uncomfortable decision into an ordinary impulse buy.
Every bargain needs both sides to forgive themselves
The seller can say they merely resell what an upstream supplier provides. The supplier says they only provide inventory. The router operator says software is neutral. The customer says they bought only one account. Each sentence is small enough for its speaker to avoid feeling like the cause. Combined, they provide the supply, inventory, tooling, revenue, and demand required for the market to persist.
Responsibility is not equal. Sellers control the information, write the terms, and collect the margin, so they carry more of it, especially when they deliberately place the Plus label on access that is not equivalent to Plus. Buyers face real price pressure and real needs, so dismissing all of them as greedy or stupid would be lazy. Explaining a choice, however, does not erase the harm that choice helps cause.
One $1.50 order does not make anyone feel complicit. Thousands of those orders are the reason another batch arrives, the sales bot keeps accepting payments, and another reseller keeps advertising. Without demand, murky supply is only a pile of credentials waiting to die. With steady demand, it becomes an industry.
I found a market, not a few spare accounts changing hands
While following shops that sell ChatGPT and Codex access, I encountered automated menus, QR and wallet payments, restocks of 100 to 500 accounts, claims about larger batches, mailboxes sold in volume, stock-management APIs, referrals, SMS rentals, and several ways of packaging credentials. These details do not all sit in one place. They are spread across shops and reseller layers, but together they look very much like a market with suppliers, inventory, and a routine for replacing dead products.
One shop alone posted 23 ChatGPT stock notices advertising 277 units in total. Under the deliberately extreme assumption that every unit sold at the listed price, their combined value would be 16,075,000 VND. I do not have completed invoices, refunds, chargebacks, or sourcing costs, so turning that number into a revenue headline would be careless. But 277 products in one sample are enough to break the image of someone casually reselling a spare account. There is a rhythm of restocking, grading, wholesale, retail, and replacement.
What convinced me was not a message boasting about "thousands of accounts." Those claims are easy to inflate. The boring details were more persuasive because they kept returning, including product codes, remaining stock, collaborator pricing, warranty windows, dead inventory, replacements, and the moment a shop filled its shelves again. Anyone can invent a large number in seconds. Maintaining the appearance of routine operations is harder.
This is no longer one person sharing an account with you. It is a market where buyers may be unable to log in, the quota may run out, or the account may disappear after a few days.
Where does a $1.50 account travel?
I have split the flow into a few pieces to make it easier to see, not because every shop has five neat departments. One person may only sell retail. Another holds the supply. Someone else operates the pool. A website may only be the storefront. The buyer usually meets the final layer and has little idea how many people handled the request before it reached a model.
An account can begin with a mailbox, phone number, payment route, trial, seat, or wholesale source. After activation, it is tested for access and remaining quota. A usable account is delivered directly or imported into a pool for a router to manage. The seller can then offer an account, slot, key, or endpoint and replace the upstream source when an account dies.
First, there has to be a supply
UPI, Apple Pay, GCash, iCloud, Gmail, and Hotmail appear in the chats almost like product categories. They may describe a mailbox, payment method, or expected lifetime. OpenAI officially supports UPI for Go and Plus in India, so seeing "UPI" and immediately calling it an exploit would be wrong. The better question is how the account was activated, how long its entitlement lasts, and why the route can be used or acquired at a scale large enough to feed multiple reseller layers.
I do not believe one machine produces every account in this market. A more realistic answer is that several sources are mixed together, from trials, promotions, regional routes, seats or workspaces to accounts bought upstream, batch-created supply, and unused quota. Stolen credentials and payment fraud can absolutely be present, but price alone is not enough to label every batch "carded." Grey markets are usually dirty in several ways at once. They are not considerate enough to use one recipe just so the story is easy to tell.
Then each account gets a health check
A seller needs to know whether an account can log in, which models it can use, how much quota remains, whether it asks for phone verification, whether its session is alive, and whether the buyer can change its details. Words such as "scan," "die," "durable," "no warranty," and "10-day warranty" look chaotic until they are placed together. Then they resemble inventory grading. An account is not merely alive or dead. It has a quality grade.
One LINUX DO post claimed that more than 30 trial accounts passing through a UPI route had fallen to eight survivors by a later update. That is a self-report, not an audited OpenAI statistic. Still, the attrition fits surprisingly well with the shops' constant cycle of filling stock, rescanning, replacing accounts, and shortening warranties. If only one of every three or four accounts lives long enough, the supply has to be much larger than the active customer count.
Packaging decides what the buyer thinks they bought
Selling a credential is the simplest model: the buyer receives an account and signs in. An activation or seat keeps the account on the buyer's side while the Plus entitlement comes through someone else. A relay hides the account completely. The buyer sees only a key and base URL, while the operator keeps OAuth tokens, sessions, or upstream API keys in the backend. All three can be advertised as ChatGPT or Codex, but their ownership, privacy, and stability are entirely different.
Finally, a router moves the mess out of sight
When a standalone account dies, the buyer notices immediately. Inside a pool, the same event may become one red line on an operator's dashboard. The router sends requests to another account, lowers the priority of a source near its quota, refreshes a token, and keeps the public endpoint unchanged. Instability does not disappear. It is moved behind a curtain where the customer cannot tell whether they just switched from account A to B or C.
9Router does not create accounts. It makes the inventory look less chaotic
I found 9Router while following the way shops talk about routers, quota, and importing accounts into pools. There is no reason to assume every shop runs the same software, but 9Router is the closest public example of this middle layer. Its capabilities help explain how an uneven account inventory can be packaged into a service that appears stable.
The 9Router README openly describes multi-account support, load balancing, automatic failover when a source runs out of quota, real-time quota tracking, and mixing subscription, cheap, and free tiers. Its architecture document places Codex CLI and other clients on one side, 9Router in the middle, and OAuth providers, API-key providers, and compatible nodes behind it. One look at that diagram makes the role clear. It is not a Plus factory. It is a dispatch desk.
That role has many ordinary uses. A developer may manage accounts they legitimately own. A team may balance authorized providers. An internal system may need failover when its primary source goes down. Round-robin is not a criminal offence. But when a router is combined with bulk-bought credentials and the resulting endpoint is sold to unrelated customers, the economics change. The operator no longer needs to deliver a complete account to every buyer. They only need to keep requests flowing.
What 9Router amplifies is not the account count but the usefulness of each account
Account A can serve one group of requests until it is nearly empty. Account B takes over. Account C stays in reserve. One source dying overnight does not have to make the customer-facing service die with it. A few hundred accounts can therefore support more sales than the number sitting in inventory. The product slowly changes from "your account" into "permission to use a pipe that still works."
This is why I disagree with both extreme stories. Saying 9Router creates cheap accounts gives it the wrong role. Saying it is unrelated because it is neutral software ignores what it actually enables. It does not manufacture supply, but it helps combine uneven supply, use leftover quota, and keep a storefront alive longer than any individual account behind it.
Cheap supply, volume, and short warranties turn attrition into profit
I cannot find a sensible business model in which a seller buys thousands of Plus subscriptions at list price, gives each customer a separate full subscription, and consistently sells them for 15,000 to 150,000 VND. If the observed prices are real and the sellers keep operating, at least part of the sentence "every buyer receives one complete first-party Plus subscription" has to be false.
Input cost may be lower because of trials, promotions, regional routes, activations, seats, or wholesale sources. An account may be resold. Multiple accounts may be pooled and shared across more customers. The product may simply carry the Plus label without being equivalent to a dedicated subscription. Money or credentials that do not belong to the seller are possibilities that must be considered, but identifying a particular batch still requires payment or takeover evidence.
One person publicly reported an input cost of about 5 RMB per account. Another said that only eight of more than 30 accounts were still alive later. These figures do not come from the books of the same shop, so I do not use them to calculate profit. They still illustrate a simple problem. If only eight out of 30 accounts survive, an initial 5 RMB has become almost 18.75 RMB per surviving account before mailboxes, SMS, labour, refunds, and accounts that die later.
High attrition can multiply the cost of apparently cheap inventory. To stay profitable, a seller needs to move it quickly, use quota before it dies, place it in a pool, sell it more than once, or shift risk to customers through an extremely short warranty. That is where the three-hour warranty fits into the economics. It does not guarantee quality. It keeps an account that dies later from becoming the shop’s loss.
“Thousands of accounts” is a number designed to make people dizzy
During my research, I encountered batches of dozens or hundreds and self-reported figures around 800 to 1,000 accounts. Elsewhere, people claim inventories of thousands of Free accounts. I think this is enough to say bulk pipelines are real. It is not enough to conclude that one seller has hundreds of thousands of fully paid, unique, simultaneously live Plus subscriptions.
A market boast may count every account ever created, including dead ones. It may count credentials delivered, with the same account appearing more than once. It may count requests through a pool or listings repeated by a bot. Whenever I see "10,000," I want to ask a question that ruins the mood. Ten thousand what?
Open Measures found more than five million Telegram posts showing signs of advertising or requesting accounts across many platforms between July 2025 and July 2026. About 429,720 mentioned AI platforms. The 15 most active accounts alone produced nearly 936,000 posts, much of it through automation. These figures show that account distribution is a very large market. They also warn against treating post count as a count of unique ChatGPT accounts.
This market does not require a secret OpenAI hack
The idea that someone hacked OpenAI and obtained infinite Plus accounts is satisfying. It turns everything into one large secret with a clear villain. The chats I read are much more ordinary. Sellers discuss mailboxes, phone verification, payment routes, quota, dead accounts, restocking, and warranty replacements. This is the language of exploiting differences and scaling them through volume, not the language of a stable backend zero-day.
OpenAI does offer trials and promotional subscriptions subject to eligibility. Referrals may provide usage credits, rate-limit resets, or temporary benefits, although they are not automatically API credits and cannot be sold or transferred. Codex limits also vary by plan, model, and task complexity. Those three mechanisms alone create several forms of quota with different values and lifetimes for people to collect, transform, and resell.
Okta Threat Intelligence's Free tokens for sale documents a broader pattern. Fraudulent registration obtains trials or credits, then relay stations and gateways turn them into endpoints advertised far below official prices. Okta does not identify the source of accounts sold by the particular shops I saw. It shows that the path from account factory to credit or quota, then through a pool and into a cheap endpoint, exists across AI services.
The buyer is gambling with more than $1.50
Thirty or forty thousand VND makes failure look harmless. If the account dies, buy another one. But the purchase price is the smallest and most visible risk. If a seller provides a credential, they once knew the email, password, and 2FA secret. If they provide a relay, prompts and responses have to pass through their machines. With Codex, a prompt may bring source code, file paths, error logs, project configuration, tokens, and work data with it.
- A previous user may have consumed most of the quota while the account still carries a Plus label.
- The buyer may lose access immediately after the three-hour warranty ends.
- Abuse by another user in the same pool may cause the entire source to be suspended.
- The gateway may log requests and responses for debugging, usage measurement, or something the customer is never told about.
- The model behind the endpoint may not match the advertised model name.
- Credentials and data may pass through several anonymous reseller layers.
OpenAI says an account is meant for the person who created it. Its Terms of Use prohibit sharing credentials, making an account available to others, and bypassing rate limits or protective measures. I am not quoting policy to pretend the real-world market disappears because a webpage says so. The terms show that this access can be revoked at any time, and that a buyer has little basis for asking OpenAI to rescue an account they were never allowed to receive.
An OpenAI-compatible endpoint does not mean the request goes directly to OpenAI. “Compatible” describes the shape of the interface, not who reads the data, where the model runs, or who owns the quota.
Even a router can forget to lock the door
A gateway holds almost everything valuable in the system, from tokens, accounts, API keys, and usage logs to request content. Combining many sources in one place makes operations easier, but it also creates one place where a bad configuration can expose the whole inventory.
9Router has had public security advisories that were fixed across releases. An authorization bypass affected versions before 0.3.75. Unprotected MCP plugin endpoints could lead to RCE, or remote code execution, from 0.4.30 to versions before 0.4.37. Another advisory described an open /v1 proxy and SSRF, or server-side request forgery, through 0.4.80, fixed in 0.4.82.
Mentioning these bugs does not make 9Router malware, nor does it mean current versions retain the old flaws. It breaks another illusion. Running a gateway is not a one-click install that can be forgotten. Operators need to update, require API keys, restrict the network, keep dashboards off the public Internet, and understand their logs. A pool lets accounts share load, but the pool itself becomes a very large single point of failure.
The remaining costs fall on ordinary users too
A pool produces traffic unlike one person. It brings many sessions, many IPs, parallel requests, quota usage near the ceiling, and immediate account replacement when a source runs out. The provider's predictable response is more phone verification, stronger device linking, account-relationship detection, tighter rate limits, and promotion revocation. Market "sweeps" may be the seller-visible side of that chase, although without OpenAI logs I cannot name the exact mechanism that suspended each batch.
The frustrating part is that anti-abuse controls rarely touch only sellers. A student with several devices, a developer changing networks, a travelling team, or a legitimate heavy user may face more verification too. The grey market keeps the profit created by gaps in the system, while the inconvenience is distributed across the user base.
OpenAI cannot simply stand outside the problem and point at its terms. When plans, quota, credits, and usage accounting are difficult to understand, ambiguity becomes inventory for resellers. When a third-party endpoint can call itself an "official Codex API" while buyers struggle to verify the model, quota source, and privacy, the provider still has work to do in product transparency, compromise warnings, and session controls.
I do not want open-source software to become the convenient villain
I have spent too much time around open-source software to accept the conclusion that multi-account support automatically makes something a hacking tool. A person can legitimately own several accounts. A company can hold several seats. A team can build an internal gateway for resources it is authorized to use. 9Router, codex2api, or a load balancer is not guilty merely because it contains a pool.
The troubling line appears when credentials are sold to unrelated people, dead inventory is continuously replaced, account replacement becomes the default warranty, the source is hidden behind an endpoint, trials or seats are commercialized beyond their purpose, and buyers cannot tell where prompts go. At that point, "software is only a tool" remains true but answers only the code question. It says nothing about where the accounts came from, what rights the operator has to them, or whose hands customer data crosses.
Flattening every tool into "malware" harms legitimate users and gives questionable operators an easy rebuttal. I would rather criticize the right thing. 9Router does not produce the supply. It makes the supply easier to combine and sell as a service. Responsibility lies in how that supply is obtained, how access is divided, and how customer data is handled.
There are things I still do not know, and I am leaving them that way
I do not know exactly what percentage of accounts comes from trials, payment routes, seats, account factories, takeovers, or payment fraud. I cannot say how many live Plus subscriptions a seller holds at one moment or which router every shop uses. Those gaps are real.
The gaps do not erase what is visible. Bots, stock counts, wholesale prices, warranties, credentials, referrals, SMS rentals, restocks, and survival grading recur across shops. Combined with router documentation, Okta's research, and the distribution scale observed by Open Measures, the broad shape is hard to miss. Supply is collected in batches, tested, packaged, placed in pools or relays, sold, and replaced when it dies.
If OpenAI later explains its trial mechanisms more clearly, changes UPI, removes a promotion, or redesigns Codex quota, I will update the sources, prices, and survival assumptions. This article is a dated snapshot, not a prophecy. It would become a joke only if the listings delivered no real access, the pools did not use upstream accounts as described, or the entire rhythm of restocking and replacement had been fabricated. Future change does not make past events disappear.
If you still want to buy, do not call it a harmless bargain
I understand why cheap accounts have buyers. A student who needs Codex for an assignment or a personal project may find the official price too high. Forty thousand VND is much easier to approve than a monthly subscription. But an affordable price does not make source code, prompts, and personal accounts less valuable. It also does not stop the buyer's money from telling the shop to order another batch.
- Is this a private account, seat, activation, or relay endpoint?
- Who controls the email address, recovery options, 2FA, and old sessions?
- Are prompts and responses logged, where are the logs, and how long are they retained?
- How are the model and quota verified beyond the seller's screenshot?
- When the account dies, does the seller replace it or refund it, and what happens to old data?
- Does the seller actually have permission to provide that seat or entitlement?
I would not send source code, tokens, customer documents, or work data through an untrusted relay. If you already use a resold credential, do not reuse a password, attach your payment method, or place personal data in it. If you operate a gateway for a legitimate purpose, update it, restrict the network, enable authentication, and inspect the logs before trusting that the word "localhost" has protected everything.
What did I compare against?
To establish the official baseline, I read OpenAI's Account Sharing Policy, Terms of Use, multi-currency billing guide, Codex usage limits, Codex Referral Promotions, and the trial FAQ. I use them as points of comparison, not as a reason to deny what happens in markets simply because OpenAI has not published an article for every abuse pattern.
The router analysis draws on the 9Router repository, its architecture document, and advisories covering authorization bypass, RCE through MCP plugins, and an open relay with SSRF. Okta Threat Intelligence and Open Measures provide wider views of account factories, relays, and the way Telegram automation magnifies apparent scale.
Technical secrets do not keep this market alive. Consent to ambiguity does
After reading everything, 39,999 VND no longer feels like the strangest detail. The price is only the result. What makes the market work is the distance between the label the customer sees and the asset behind it. "Plus" may be a trial. "API" may be a relay. "One month" may be the advertised period for an account that survives three days. "Warranty" may mean the right to receive another credential from the same uncertain inventory. "Unlimited" may mean the router keeps changing sources until the pool is empty.
Sellers need that ambiguity to protect the margin and avoid promising too much. Buyers accept it because the price is low, the need is immediate, and each person hopes to be one of the lucky ones. 9Router does not create this bargain, but it can make the bargain easier to scale. Uneven accounts become one continuous-looking endpoint while backend disorder temporarily disappears from the customer's view.
The persistence of unethical sellers is therefore not mysterious. They survive because risk has been transferred to someone else, sales have been automated, responsibility has been split across layers, and there is always another buyer who decides that losing $1.50 would not matter. The crowd is not necessarily naive either. Many buyers know the product is questionable, but immediate utility and the tiny price let them grant themselves permission to ignore the rest.
Sellers cannot keep ordering inventory without buyers. Buyers cannot pay so little unless someone behind the product is absorbing the cost, losing access, or having an entitlement taken and resold.
That is why I do not want to end with "every cheap account is hacked" or "9Router is malware." Those lines sound forceful, but they are too easy and miss the important part. The thing worth criticizing is a market that turns opacity into a product. Sellers keep the margin. Customers receive dead accounts and data risk. Ordinary users receive more anti-abuse controls. Every small transaction pays for the cycle to begin again.
You made it to the end The short version
Buyers often do not see it as a complete month of Plus. They want immediate access for an urgent task, see $1.50 as an acceptable loss, and treat three hours as enough time to log in, test the product, and hope it survives longer.
They acquire cheap supply, sell quickly, share quota through pools, repackage access in several forms, and limit warranties. When an account dies after three hours, failure after the deadline becomes the customer’s loss instead of the shop’s cost.
One order looks insignificant, but thousands of similar decisions pay for the next account batch, sales bot, and reseller layer. Economic constraints explain the buyer's choice, but do not stop that demand from financing supply.
9Router does not create accounts. It shows how a gateway can combine accounts, track quota, and replace a failing source. That mechanism turns unstable inventory into a continuous-looking endpoint and lets an operator extract more usable quota.
- NhanAZ - 10.08.2026