Try your luck?
© 2026 ReClaude. All rights reserved
2-seat and 4-seat plans split the quota by share. Works with Codex desktop and CLI. How do we guarantee others won't use up your quota? This page explains the mechanism—no sign-in required.
We split Pro 20x and Pro 5x account quota into four equal shares. Whichever tier you buy, you get a fixed number of shares—the ratio is hard-coded in the plan and won't change based on who joins first or acts faster.
This tier gets 1/4, or 25% of the total account quota.
Every request you send is recorded only against your own ledger. No matter how heavily others in the same share use it, only their own slot is deducted—not yours.
Four people share one account, each with an equal-height slot.
This is the step most people ask about. The official system reports only one number for the whole account, but the bar on your dashboard shows your own share—the difference is just a division.
Reading unchanged—no one touched it during this period
One account's quota is the whole cake. If you buy a 4-person ride, your share is the 25% on the left.
Many people think quota works like a phone plan that resets on schedule. It doesn't. OpenAI's official quota already uses rolling windows: Plus has a 5-hour window and a 7-day window, while Pro has only a 7-day window. Usage from 5 hours ago slides out of the 5-hour window after 5 hours, freeing the quota. Window rules follow official adjustments.
We fully follow the official two windows and add no other restrictions.
This is usage in the official 5-hour window. Each cell is one hour; the rightmost cell is what was just used.
Many shared plans in the market calculate it this way: first estimate a USD price for Pro 20x, then divide that USD amount by 4 among four users. The problem is the first step.
The converted amount is entirely the seller’s word, and you cannot verify it. A lower valuation benefits them—they can sell more shares of the same account.
The official system measures “what percentage of the 5-hour/weekly window has been used,” and we split by that percentage. 1/4 means 1/4 of the account’s capacity.
The key point: the official system has never provided USD pricing for Pro 20x or Pro 5x.
The official system measures it as a percentage of the quota window; that’s a different product and pricing model from token-billed APIs, so they can’t be converted. Any claim about “how many dollars Pro 20x is worth” is just someone’s estimate. Our backend also shows a USD reference value, but it’s only an approximation—the request details state clearly: this value is for reference only, and plan users are not billed by this amount. What actually determines how much you can use is that percentage.
OpenAI itself has been a bit unstable lately: disconnects, queues, and inconsistent speed. People who reach out often assume it's our issue, but at the same time, customers with official subscriptions run into the same problems. But our word isn't proof. That's why we added Direct Mode — requests don't go through us, so there's one less variable.
To use Codex from mainland China, there are basically three paths: subscribe to the official service yourself, use a relay/reverse proxy, or use a carpool direct connection. The difference isn't the price tag—first look at how many hops your request passes through on each path, and where it gets opened.
The connection goes straight through, and no one can decrypt it. But you must handle the middle hop yourself.
The connection is opened and rewritten on the server, then a new request is sent out.
Same shape as ①, except we handle the middle hop.
If your VPN drops or switches nodes, Codex will disconnect and reconnect.
Egress IP decides whether you get the full model: we tested 15 exit routes, and all 8 US ones were downgraded. Most VPN nodes land in US data centers.
Requires a foreign-currency card and paying the official list price.
Three extra hops, each a possible failure point and source of latency.
Requests are rewritten: the model may be swapped for a cheaper one, long responses get cut off, and web access and image generation disappear.
Conversation content passes through a third-party server.
Only one middle hop—same as subscribing to the official service yourself, without needing your own VPN.
We pick egress routes from real-world tests and switch automatically if a route is downgraded.
Requests pass through unchanged, so plugins and other native features work normally; content doesn’t pass through us.
Just update the client or CLI to the latest version; no configuration changes are needed.
Want to check? Look at the request type in the console’s “History” — if it shows “Direct,” it didn’t pass through us.
Only the ReClaude client and CLI are supported. We can’t support pasting the API key into other tools — direct connections require a process running on your machine to establish the connection.
The limits of Shared Direct, stated honestly
The account is shared among several people: it’s cheap because it’s shared, but your quota is still your own, and rate limits apply once you use it up. The egress is our fixed route — each person connects from their own network, so upstream sees “one account logged in from several cities at once.”
Your own usage can be checked item by item in the console’s “History,” including how much quota each request deducted.
Besides our own accounting, we also read the official usage for the entire account and reconcile the two.
Any mismatched portion (for example, usage elsewhere on the account) is recorded separately as “Unattributed” and is not charged to anyone.
Shared plans are not “unlimited,” and no such thing exists. You can use only your own share. When you use up your share, you’ll see a rate-limit notice. No need to contact support—once the window slides, your quota returns on its own, usually within a few hours. Conversely, if another user in the same shared plan is rate-limited, you can keep using yours as normal, unaffected. If your usage consistently runs at the cap, this tier isn’t enough for you; switching to a dedicated plan is more cost-effective than a shared plan.