Breaking news, OpenAI is officially on duty 24 hours a day! GPT boldly launches the most expensive package at $500

The Fort Mason venue in San Francisco was ablaze with excitement at the OpenAI Developer Conference!

Just now, Ultraman took the stage at DevDay and unleashed 20 major announcements in a single hour.

A brand new species called dots, an AI that is online 24/7, has officially been born.

Unlike ChatGPT, it is OpenAI's first independent cloud computer that can run code and monitor projects for humans anytime, anywhere, serving as a "Resident Agent."

Even more shocking, Ultraman introduced the next-generation flagship GPT-6.1 Sol, featuring top-tier reasoning and coding capabilities at the Astra level, with prices slashed by 80%.

The ultrafast mode was launched simultaneously, boasting a terrifying generation speed of 300 tokens per second, which is 8 times that of the standard version. On-site, Ultraman reopened the $200 Pro subscription and introduced the "Noble Package" for the first time—Pro 500.

Its usage is 25 times that of ChatGPT Plus, allowing access to ultrafast mode without a 5-hour limit.

In addition, at the conference, OpenAI also launched ChatGPT Space, Agents API, Codex cloud version… the amount of information was overwhelming.

At the very end, "Godfather" Tibo pressed a reset button and performed a one-click restart for everyone on-site.

Some netizens commented that the most exciting part of the entire conference was this segment.

Dots is here

OpenAI's first 24/7 always-online AI

The biggest star of this conference is called dots.

It is not just another chatbot from OpenAI, nor is it a new feature of ChatGPT.

As previously leaked, dots is an all-weather 24/7 online Agent powered by GPT-6 Astra.

It not only has its own cloud computer and browser but also seamlessly integrates with over 4000 tools, directly entering Slack and Teams.

Give it a goal, and it will take care of it on its own without needing you to teach it step by step. Even with the screen off, dots continues to run tasks silently in the background 24 hours a day.

Without logging into ChatGPT, dots can start working for you anytime, anywhere.

The first meeting process feels quite ceremonial.

When you open ChatGPT, your dot will greet you, "Hi, I am your dot, just reporting to ChatGPT. From now on, leave all the tasks that come to me."

On-site, OpenAI researchers named it Alfred.

From the moment it was activated, Alfred took over all fragmented tasks in the background------

• Automated website building and launching

Based on the approved design drafts, Alfred independently pushed for website updates in the background, only prompting for human decisions (like replacing the first and third images) when necessary, and immediately completing the launch upon confirmation.

• Cross-platform data tracking and board reports

It autonomously monitored user feedback data in Teams, directly synchronizing updates to the board report PPT, even accurately capturing core metrics------

Product usability increased, with annual revenue growing by 51%.

• Long-term project restructuring

It captured the information "App needs to migrate" from Google Meet meeting minutes, automatically checking processes and drafting architecture diagrams.

There’s also a real case from within OpenAI: an employee forgot to invoice a publishing agency, and his dot discovered it, automatically prepared the invoice, and sent a message asking, "Should I send it out?"

He clicked confirm, and the task was completed.

This is the fundamental logic shift of Dots: it is not a tool; it is a colleague.

You have hired a digital employee who never takes a day off, never forgets things, and is online 24 hours a day.

Moreover, it will understand you better the more you use it—your preferences, your standards, and what "good" looks like in your mind, it is learning all of it.

Of course, Ultraman did not shy away from security issues. Each dot runs on an independent cloud computer, with personal computers and data isolated by default. When it uses saved passwords, it does not expose the passwords to the model.

When it is not executing tasks, it only conducts "active research" in the background with read-only permissions, without sending messages or modifying any content.

Ultraman said a phrase on stage that is worth pondering, "You are not using a tool; you are building a team."

Starting today, Pro, Business Premium, and Enterprise users can create their first dot, and conversations do not count against ChatGPT usage limits.

ChatGPT Space, workflows completely restructured

If dots are your digital employees, then ChatGPT Space is the office prepared for these employees and their human colleagues.

This release is quite interesting. Over the past year, all the Silicon Valley giants have been stuffing AI into documents—Google Docs with AI, Notion with AI, Microsoft Office with Copilot.

But OpenAI turned it around: it directly created an "AI-native" collaboration space.

In Space, you can collaborate with colleagues and each of your dots on the same page.

You can write plans, conduct research, generate charts, and perform data analysis on the page.

You can pull dots in and ask them like you would a colleague: "Help me turn this into a bar chart." The chart is generated immediately, and you can filter and view data points in real-time.

More importantly, Space ensures that the work of dots is no longer "hidden in someone's chat history." Everyone on the team and all dots share the same knowledge base, and project context will not be interrupted just because a person changes.

In Ultraman's own words, "The pages and documents in Space are all together, like a cloud drive. But Space is alive."

GPT-6.1 Sol makes a stunning debut

Astra-level intelligence at 1/5 the price

Originally, there were not many expectations for the new model release at DevDay.

The result was quite surprising; Ultraman directly unveiled the next-generation flagship GPT-6.1 Sol, receiving thunderous applause on-site.

This is another major release from OpenAI just a week after the GPT-6 Sol launch.

GPT-6.1 Sol is an extremely powerful model, achieving performance that matches Astra in the software engineering DeepSWE v1.1 test.

Moreover, in the long-term business automation evaluation Automation Bench and scientific reasoning Terminal-Bench Science, it directly surpassed Opus 5.5.

In the operation of completing long process tasks on a computer (OSWorld 2.0), 6.1 Sol is only 2.1 percentage points away from Astra, with costs about one-seventh.

In terms of pricing, GPT -6.1 Sol once again shatters the threshold.

Input is $2 per million tokens, output is $10. Compared to Astra's $10 and $50, it is only one-fifth.

Even more exaggerated is the context cache, which costs only $0.10 per million tokens, with discounts up to 95% compared to standard input.

On-site, Ultraman also launched the Ultrafast inference mode, achieving speeds 8 times that of the standard mode, approximately 300 tokens per second.

In the live demonstration, for the same task of "building a rocket," the rocket on the Ultrafast side had already launched, while the standard mode was still assembling the model.

Currently, Ultrafast initially only supports Astra, but will later be opened to Sol.

To match such terrifying call frequencies, the previously paused Pro 200 membership has been fully restarted.

Not only that, OpenAI has further pushed the Pro 500 tier, with 25 times the calling quota of Plus and native usage rights for Ultrafast, completely unlocking productivity limits.

Codex Moves to the Cloud

Write code without being tied to a computer

DevDay is after all the developers' main stage, and Codex has not been idle.

The biggest change is that Codex has moved entirely to the cloud.

The development environment can be pre-configured and reused repeatedly, with permissions set once. When you are outside, you can control it remotely with your phone.

The command line version Codex CLI has also been updated, allowing direct voice commands and mid-course direction changes.

The new /agents view can monitor several sub-tasks simultaneously, and it won't get chaotic even when left open for a long time.

The desktop version of ChatGPT has added code review.

When you step away from the computer, it will automatically review the changes, and the review comments can be pushed back directly to GitHub or GitLab.

Codex Security Cloud is specifically designed to catch security vulnerabilities.

It can scan entire GitHub repositories, reviewing each new submission, self-checking, deduplicating, and preparing repair plans for your review, with OpenAI's dedicated cybersecurity model integrated by default.

There are also significant movements on the API side.

The underlying technology that drives dots has been directly packaged into the Agents API.

Computer operations, multi-agent collaboration, tool searches, and context compression can all be directly called by developers, essentially allowing them to create their own dots.

Another new interface, Decisions API, is specifically designed to handle decisions like "quickly picking one from several fixed options," such as categorizing an email or deciding which button the agent should click next.

Editors who tried it in advance ran a computer operation playback, making 76 correct choices out of 78, with a response time of only 230 milliseconds.

However, in another set of dialogue classification tests, it tied with competing models in accuracy but was slower.

Tibo also shared some numbers. The token traffic of OpenAI's core interface Responses API has increased 100 times compared to the same day last year, with availability remaining at 99.9%.

Personal Agent Battle

OpenAI Reveals All Cards

Tonight's more than 20 releases, each one could be written about individually.

Looking at them together, OpenAI has essentially done one thing: transformed ChatGPT from a place where you ask questions into a place that does work for you.

The urgency is due to competitors already racing ahead.

Meta's Muse has surpassed 3 million downloads in just three weeks, available for free but with a weekly usage limit, and is rapidly expanding into small business scenarios. Google's Spark has long been working for users 24/7 in the cloud, and SpaceXAI has also launched Grok Bot.

Personal agents have become the next entry point that every tech giant wants to seize.

The uniqueness of dots lies in its working method: before you even speak, it has already placed the results in front of you. It doesn't ask "What do you need me to do?" It directly says, "I’m done."

At the end of his speech, Ultraman said, "Some compare AI to a new industrial revolution. I don't quite like that analogy. If done right, it resembles a new renaissance more."

The core of the renaissance is to give people more time to do creative and meaningful things.

If dots can really help you handle those boring, repetitive, attention-consuming tasks—what do you plan to do with the time you save?

This launch event has released a clear signal: AI is transitioning from a "tool" to a "colleague," from "you using it" to "it working for you."

The transformation has already occurred. Starting today.

Your dots are already waiting for you.

References:

https://www.youtube.com/watch?v=Fls_onRviPM%20dots%EF%BC%9A%20

https://openai.com/index/introducing-dots/%20

https://openai.com/index/introducing-gpt-6-1-sol%20