新智元报道
还是OpenAI会玩！
前脚刚把Pro额度翻倍，大家还没高兴两天，转头就宣布——
从今年12月开始，ChatGPT和Codex的「使用额度」将正式合并！
划重点！！！
不管你是Plus、Pro还是Business用户，以后在ChatGPT里深度提问，和在Codex里跑代码，耗的都是同一个池子里的额度。
就连底层的计费逻辑，也一同变了——「全部统一按Token算」。
几乎同一时间，OpenAI还上线了严厉的「设备绑号上限」。一台设备若是绑太多付费账号，直接封锁。
如今，全网开发者都吵翻了，所有人更关心，同样一笔月费，究竟还能干多少活？
就在6小时前，OpenAI官方支持页面悄悄上线了一篇新文档——
「理解ChatGPT与Codex使用限制」。把官方文档拆开看，核心就四点：
第一，额度合池。第二，按Token计量。第三，Instant不扣。第四，用完只有三条路。
这一次大改，是为了将Work「工作空间」的智能体能力，直接塞进ChatGPT主界面。
也就是从今年12月起，Plus、Pro和Business套餐的使用配额，将在ChatGPT和Codex之间全面共享。
以前，大家习惯的可能是「几小时内能发多少条消息」的粗放式计算。到了12月，这套标尺彻底换成了「按Token计量」。
官方对此给出了明确界定，你消耗多少配额，完全取决于选择的模型，以及让它干的活儿有多重。
调用更高级的模型、执行更复杂的推理和长代码任务，通常会消耗掉更多的配额。
好在OpenAI还算留了条退路，日常「即时聊天」依然无限量供应，不占额度随便聊。
至于额度全部用完之后，摆在面前的有三条路：
坐等刷新重置；
掏钱升级到更贵的订阅档位；
直接氪金额外的Credits「续命」。
合并之后，Plus、Pro、Business各档到底能拿到多少Token总盘子？不同模型之间的消耗比例怎么折算？
目前，OpenAI并未公布具体的Token配额数字与折算表。没有明细，大家心里肯定打鼓：这不就是变相缩水吗？
不过，他们在文档里特意打了一剂「强心针」，承诺原话大概是这么说的：
绝大部分人被卡额度、等重置的频率，肯定不会比现在更多。
话虽这么说，但在各大开发者社区里，重度用户们已经在背地里疯狂算细账了。
开发者们的担忧非常实在：以前ChatGPT的深度推理和Codex的Agent跑代码各算各的账，井水不犯河水。
到了12月，两边共用一口锅吃饭。
白天用ChatGPT做了一堆深度的文档分析、跑了几个复杂的Agent任务，晚上打开桌面端准备用Codex重构项目时，系统突然弹窗提示额度告罄——
看着满屏的「额度焦虑」，让不少全栈工程师直呼「得悠着点用了」。
Anthropic的Claude，其实早就这么干了。
Pro和Max套餐的额度，在Claude聊天和Claude Code之间共享，两边所有活动都计入同一个上限。
白天在网页上聊多了，晚上终端里的Claude Code就少了。而且就在昨天，Anthropic刚刚动了自家套餐：
原来的Max 5x改名Max，月费仍是100美元；原来的Max 20x改名Ultra，月费仍是200美元。
最大的变化是，通过验证的Ultra账户，取消了5小时用量限制，但每周总限额不变。
两家放在一起，对比立刻明了：
就在10月30日，原本按照之前的过渡计划，Pro用户的促销额度倍数即将面临回调。
当时整个开发者社区怨声载道，不少人甚至放话要退订。
结果，OpenAI的Codex负责人Tibo突然高调开启「宠粉模式」，一口气甩出更新：
Pro 100/200/500全档位和Business Premium，额度即日起翻倍。
就在人们还沉浸在「额度翻倍」的狂喜中时，一封邮件顺理成章地引出了真正的后手——
Chat与Work标签页即将合体，ChatGPT与Codex的配额也将在未来几周内彻底打通。
邮件中透露，接下来几周内，OpenAI会先在小范围用户中测试「Chat与Work合一」的全新界面。
以后不需要在聊天框和工作台之间来回横跳，直接在同一个地方就能把文件、插件拉进来，让ChatGPT帮你建网站、做PPT、在网页上自动跑任务，而人类只需要在旁边审阅和指导。
至于Codex，则会继续在ChatGPT桌面端App里保留独立入口。
早在几天前，有人便发现ChatGPT入口更新，原来顶部的「聊天｜工作」切换开关消失。
这个今年7月上线的新功能，在OpenAI内部看来并不理想——
总裁Brockman在采访中承认，新版应用有点乱，并承诺到年底就不会再有单独的Work标签。
OpenAI这波操作，实际上逻辑前后相承——
10月底先把池子扩容一倍，平息社区怒火；11月测试界面大一统，培养人们高频调用Agent的习惯；到了12月，正式把ChatGPT和Codex关进同一个Token计费池。
看着是挤牙膏，其实人家早就把订阅制这套底层逻辑给彻底重构了。
就在宣布额度池合并的同一周，OpenAI反手就把「多开叠号」这条路给焊死了。
OpenAI悄悄上线了一项针对「设备限制」的硬核新规：
一台设备能够登录使用的付费账号数量，现在有了严格的硬性上限！
- 受影响范围：涵盖Go、Plus，以及Pro的100 / 200 / 500全系个人付费档位。
- 豁免范围：Business、Enterprise、Edu以及免费（Free）账号不计入这一设备上限。
一旦你在一台设备上绑定的付费号触达上限，登录时就会直接撞上冷冰冰的弹窗提示——Device limit reached（已达设备上限），或者在使用中途被强制踢下线。
哪怕你的订阅付款已经扣费成功，也照样登不上去。
最绝的是，防卡Bug机制千万别以为「退出A账号，就能腾出坑位登B账号」。官方文档明确指出，登出已有账号并不会为你释放新的设备名额！
很多重度依赖Codex写代码的极客或小作坊，之前为了突破单账号的额度瓶颈，又不愿意直接买官方昂贵的加量包，往往会「多开叠号」。
也就是在一台主力开发机上，同时登录好几个20美元的Plus号或者低档位Pro号，这个号额度跑光了，秒切下一个号继续肝。
如今，这条路再也行不通了。
过去这段时间，「义父」Tibo最出圈的操作，就是隔三差五给大家免费重置额度。
用户一喊不够用，他就出来按一下Reset，评论区一片欢呼。
但「毕业」两个字，已经把话说透了：靠手动Reset哄用户的日子，要翻篇了。
取而代之的，是一套更冷静的规则——每个Token都有价，每次调用都记账。
订阅费买到的，是一个按量消耗的「算力钱包」。从「随便用」到「精打细算地用」，AI订阅正在迎来一次底层切换。
对于重度开发者来说，每一次调用高级模型前，心里都得拨一下算盘。
12月合池之后，你打算怎么应对？
参考资料：
https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits?utm_source=chatgpt.com
编辑：桃子
不喜欢
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
New Ji-won.
Or is Openai playing!
The front foot just doubled Proquota, and before you were happy, you turned around and announced...
Starting in December this year, ChatGPT and Codex's "Use quota" will be formally merged!
Focus!
Whether you're users of Plus, Pro or Business, then ask questions in ChatGPT and run codes in Codex, it's all quota in the same pool.
Even at the bottom of the billing logic, it's changed - "All of them on Token."
At almost the same time, Open AI also went online with a severe "ceiling of equipment." If one equipment is tied to too many paid accounts, it is blocked directly.
Today, the whole network of developers is out of their minds, and everyone is more concerned about the same monthly fee.
Just six hours ago, Open AI's official support page came on line with a new document --
"Understanding ChatGPT and Codex's restrictions." Open up the official files.
First, quota pool. Second, measured by Token. Third, Instant doesn't. Fourth, there are only three ways out.
This major change was made to insert the AI agent capability of the Workspace into ChatGPT interface directly.
Since December this year, quotas for the use of the Plus, Pro and Business packages will be fully shared between ChatGPT and Codex.
In the past, you might have been used to "how many messages can be sent in a few hours". By December, the ruler was completely replaced by "measured by Token".
The official definition of this is clearly defined, and the amount of quota you consume depends entirely on the model you choose and the amount of work you do.
Calling for more advanced models, more complex reasoning and long code tasks usually consumes more quotas.
It's a good thing Open AI still has a retreat, and daily "instant chat" is still available in unlimited quantities, instead of quota.
As for quota, there are three ways forward:
Waiting to refresh and reset;
Upgrade to more expensive subscription slots;
In addition to Kryptonian Kryptonium's additional Credits, it's "renewed."
How much of Token's total is available in the Plus, Pro and Business files after the merger? How does the consumption ratio between models translate?
At present, Open AI does not publish specific Token quota figures and commutation tables. Without details, the drums must be in our hearts: isn't this just a constriction?
However, they deliberately put a "strength needle" in the file, which is probably what they say:
The vast majority of people have been replaced with card quota, and so on, certainly not more frequently.
That being said, in the communities of the major developers, the heavy users have gone crazy about the details.
The concerns of the developers are very real: the deep reasoning of ChatGPT and the Agent code of Codex are all accounted for separately, and the wells are free from the river.
By December, both sides shared a pot for dinner.
During the day, ChatGPT did a series of deep document analyses, ran several complex Agent missions, opened the desktop at night to remodel the project with Codex, and suddenly the system bullet window tip quota was exhausted --
The quota is a full screen of "anxieties", which has led a number of all the engineers to say, "Take it easy."
Claude of Anthropic, did it already.
quota, a Pro and Max package, is shared between Claude Chat and Claude Code, and all activities on both sides are subject to the same ceiling.
Too much talk on the web during the day and less Claude Code at the end of the night. And just yesterday, Anthropic just had his own home set:
The former Max 5x name was changed to Max and the monthly fee remained $100; the former Max 20x name was changed to Ultra and the monthly fee remained $200.
The most significant change was the removal of the 5-hour limit from the certified Ultra account, but the overall weekly limit remained unchanged.
The two families were put together and the comparison immediately became clear:
Just on 30 October, under the previous transition plan, Pro user promotion quota is about to be reversed.
At the time, the entire community of developers had complained, and many had even spoken out.
As a result, Tibo, the Codex chief of Open AI, suddenly launched the Poppy Model with a high-profile update:
Pro 100/200/500 full slot and Business Premium, quota doubled as of that date.
And just as people were immersed in the ecstasy of "quota Double," an e-mail logically drew the real backhand.
Chat and Work labels are about to merge, and ChatGPT and Codex quotas will be fully accessible in the coming weeks.
According to the email, in the next few weeks, Open AI will first test the whole new interface between Chat and Work among a small range of users.
There is no need to jump between chat boxes and workstations, so that files, plugins can be pulled in directly from the same place, and ChatGPT can help you build a website, do a PPT, run a job on a web page, and humans need only read and guide.
In the case of Codex, the independent access will continue to be maintained at the ChatGPT desktopend App.
The ChatGPT portal was updated a few days ago, and the original chat shop on the top has disappeared.
This new feature that went online in July this year doesn't look ideal within Open AI.
In an interview, Managing Director Brockman acknowledged that the new version was a little messy and promised that no separate work labels would be available by the end of the year.
The Open AI operation, which is actually logical--
By the end of October, the pool had been doubled to calm community anger; in November, the testing interface had been consolidated to foster high-frequency calls for Agent; and in December, ChatGPT and Codex had been formally locked into the same Token billing pool.
It looks like toothpaste, but they've already completely re-engineered the bottom logic of subscription.
The same week that announced the merger of the quota pool, Open AI's backhands welded the "More Monuments" to death.
Open AI quietly went online with a new hard-core rule against "equipment restrictions":
The number of paid accounts a device can log in now has a strict hard ceiling!
- Impact range: covers Go, Plus and Pro 100/200/500 all individual paid slots.
- Exemptions: Businss, Enterprise, Edu and Free account numbers are excluded from this equipment ceiling.
Once you reach the upper limit of paid, which is bound on a device, you will enter directly into the cold-free bullet window hint - the Device limit reached, or you will be forced to drop off the line in the middle of use.
Even if your subscription payments have been successful, they will not be available.
Most importantly, the card bug mechanism must not think that "the exit from account A will free the pit to account B". Official documents make it clear that posting an existing account number does not release new equipment slots for you!
Many of the hyperbolic or small workshops that rely heavily on Codex's code-writing codes are often “multiplied” in order to break the quota bottleneck of a single account and are reluctant to buy official, expensive packs.
That is, it's on a main power developer, with several $20 Plus or low-end Pro, which ran out of quota and went next.
This road is no longer working.
In the past period, the most popular operation of Father Tibo has been to reset quota free of charge.
As soon as the user shouted, he came out and pressed Reet, cheering the comment area.
But the word "graduation" is clear: the day of the hands-on Restet is about to turn.
Instead, it is a cooler set of rules — every Token has a price and every call is recorded.
The subscription fee is a "calculative wallet" in volume consumption. From "Approximate" to "Approximately" to "Application," the AI Subscription is receiving a bottom switch.
For heavy developers, every time an advanced model is called, you have to dial your mind.
What are you gonna do after December?
References:
https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits?utm_source=chatgpt.com
Edit: Peaches
I don't like it.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.