Just now, OpenAI's mysterious "o" code has been revealed!

It's the first time in the history of the world.

OpenAI's next medal has been revealed ahead of time!
It is rumored that OpenAI will launch a personal, 24-hour online AI assistant on September 29th, Developer's Day.
It's name is just one letter, o!
Today, developers Chetaslua have officially excavated the ironclad in the ChatGPT core code:

Lowercase o is the assistant's product name.

Fast Mode is available in 63 languages, multi-smart body sync signage is complete, and even the back of an independent mailbox is already logical.
What's more, it will be downloaded directly to all Pro subscribers.
Just recently, an official picture of OpenAI developers has again suggested that it is about to go online.

This time around on Developer Day, OpenAI is going straight to the table.

OpenAI Mysteries is hereAI is always online
The launch of OpenAI is definitely the highlight of this year's launch.
This divine prophecy began a few days ago. On the afternoon of September 25th, time in the western United States, the Codex had a failure.
OpenAI's "Godfather" Tibo posted a message: "No", an hour after the service was restored, he tweeted: "O yes... we're back", announcing the reset quota.

Almost at the same time, several clues emerged.
Some Pro users saw a line on the ChatGPT upgrade page: o, your always-on assistant ((o, your all-weather assistant)), which disappeared hours later.
Not only that, but there is also a note for translators in the ChatGPT end-code, which is the name of the assistant's product.
Also, in the internal configuration, "o" is written as the display name, and there is a postcard on the back: "-o".

Additionally, the OpenAI developer account that "We're 72 hours away from DevDay, we've been building something, it's time to get to work" is basically correct.
OpenAI is going to launch a 24-hour online smart body on DevDay for you to keep working.
The information that can now be confirmed is that its location is: long-term online, continuous operation.
The $100 Pro Lite, the $200 Pro, and the $500 Pro Max, which are not yet available, are available in three sizes: $20 Plus and $20 Codex Plus.

In addition, Codex Plus users are explicitly excluded. This means that "o" and code assistants are two product lines, with OpenAI intentionally distinguishing between "AI that writes code" and "AI that works for you".

Open the underlying codeO "long" came out.
Based on the underlying configurations of reverse engineering, "o" has been completely reconstructed from interactive logic to executive architecture.
From the core details currently leaked, it is possible to see in advance its four main killers.

Always-on (sleeps and wakes up)

At this point, you're using ChatGPT, which is essentially a "roundup game".
You type in a line of instructions, it spits out a string of text, you don't answer, and it falls asleep; once the page is overdue or the task is too long, the window is automatically interrupted.
The code "o" is reportedly a reference to the internal project "Aeon" (meaning "eternal"), whose bottom is driven by a variable that reinforces long-range tasks.

At the bottom of it is a durable cloud hosting sandbox ("Hosted Sandbox").
You give it a complex business objective, and it can run for hours, days or even weeks on a cloud background.
The environment itself, the fault itself, the data itself, until the task completely runs out of the loop. You turn off the computer to sleep, and it'll keep you up all night in the cloud.

Fast Mode: Wait for long chains to be pressed to milliseconds

There is also a "Fast Mode" (extremely fast mode) mentioned repeatedly in the code.
A lot of people don't understand: Big Models are already producing text fast enough, so why do we have to create an ultra-fast model?
It's easy to see that when AI switches from a single response to autonomous thinking, a single task is often behind tens of hundreds of "CoT derivatives → tool calls → results returns → error retests".
Typical models emit 30 tokens per second, and a complex process can take up to 20 minutes to complete.
Rumour has it that o has access to an ultra-high logic computing architecture, even using Fast Mode, a dedicated high concurrency chip cluster, to increase the speed of token generation several times.

Multi-Agent shared a message board: pull a group,AII'm going to start my own morning session.

One person's energy is limited, and so is an AI.
But in an "o" world, OpenAI has directly moved the "corporate architecture" into the system.
The "Shared message board for agents" appeared in the code, which completely broke the soldier's war:

Dominant Agent PM: Responsible for breaking down macro targets into multiple sub-tasks;
Researchers in Search Agency: Interface-wide literacy;
Code Agent (Engineer): Run in sandbox simulation and script verification;
As a result, there is a lot of uncertainty about the outcome.

Humans only need to be the boss and write the delivery standards and acceptance conditions at the top.
The rest of the time, the AI employees at the bottom of the screen meet, question each other, self-promote themselves in a shared board.

After the exclusive mailbox, the "o" button: the official postal code is on the rise

One of the biggest surprises was the "o" email processing logic in the configuration.
If "o" has an independent email receiving capability, it means that it has a "native external communication input" in the true sense.
The team copies it as it emails customers, and it automatically analyzes the attachments and syncs them to the dashboard; the downstream vendor sends an offer form, and it automatically compares the historical data, generates evaluations, and answers you with confirmation.
It began to have a communication identity that was perfectly consistent with human employees.
Today's ChatGPT is not a 24-hour employee.Many people may be confused as to whether the current ChatGPT is not strong enough.
What's the disruptive difference between writing code, doing reporting, and doing what's called an "o" living thing?
Let's put it this way, the gap between the two is essentially a species crossing between "computers" and "career managers".
If you look at this comparison table, you can see what is called a change in the quality of the product:

Over the past two years, the vast majority of people have gradually lost their sense of freshness about AI.
You ask a question, and it's written in the dictionary with a whistle, but at the end you turn off the page, and you have to check the data yourself, and you have to fill out Excel yourself, and you have to write a good code, and you have to manually deploy the error.
What everyone really needs is a "performer who will be silent for me after I've done my job".
That's why OpenAI has to do "o".
From "Chat" to "Do" to "Do" is the only meaning of living in an intelligent being.

Next week.Get ready to meet your AI colleagues.
In fact, this time, OpenAI came out too late.

Earlier this month, SpaceXAI's Grock Bot was officially launched in Moscow.
Its core design is extremely clever: each bot has its own exclusive cloud computer, logs on to existing human tools, and even connects to the computer, it never stops.
Many bots can also send each other messages and collaborate in group chats.
Meta's Muse is also on the rise. He called it a "personal agent working for you to understand your goals 24/7. "
During the first 13 days of its release, the download volume of Muse pushed Claude and Grok to the top of the App Store.

For more than a month, the track was crowded, and OpenAI's o was forced to roll.
The signal released is that the competitive focus of the next phase of OpenAI is likely to be replaced by a simpler question:

How long can an AI live for people every day?

On September 29, OpenAI Developer Day is about to unveil screenshots.
Get ready to meet the first tireless, sleepless AI colleague of your life.