OpenAI Dots: AI Agents That NEVER Stop Working?!
Publicado no IA Dots ·
Agentes de IAAI SignalDotsidioma-enInglêsMuseyt-azwHmTDp7pE
Vídeo em Inglês, publicado pelo canal AI WITH Rithesh.
Video transcript
Transcript from available captions; it may contain errors.
OpenAI just gave everyone on Pro a co-worker that never logs off. It's called a dot. It has its own computer, it reads your apps, and it starts working before you ask. In this video, I'll cover what a dot actually is, what it can do, what it costs once you read the fine print, what early testers found, and why the launch day was a little messy. So, what is it? A dot is an always-on agent running on GPT-6 Astra. The key difference from ChatGPT is simple. It isn't waiting in a chat box. Each dot gets its own cloud computer with a browser, and it's a Linux machine that even comes with apps like Blender preloaded. It plugs into the same 4,000 plus apps ChatGPT already connects to, things like Slack and Google Drive, and it keeps going around the clock on several projects at once. You talk to it in ChatGPT on desktop, web, or your phone. You can also message it in Slack and Teams, or talk to it by voice. Texting is coming later. You set it up on desktop first, you give it a name, and it gets a handle, like @yourname-dot. Yes, you can name your agent. People are already posting theirs. Here's the part that's actually new. When you haven't given it anything to do, a dot does what OpenAI calls proactive research. It looks through your connected apps, read-only, and comes back with things you might want done. One early tester woke up to a canceled train. Before he typed a word, his dot had read the cancellation email, checked his ticket was valid on an earlier train, and found that earlier train was running late. He made it. OpenAI's own example is smaller, but it's the same idea. A dot noticed a tester forgot to send an invoice, prepared it, and sent it once he said yes. A researcher at OpenAI said his dot tested over a weekend found about $500 a year in recurring charges he could cut, then handled a whole customer service text chat for him. A reviewer who used one for days said his dot warned him that a flight change would clash with a Slack meeting request he hadn't even read yet. These are single experiences, not measurements, but they show the pitch clearly. You stop managing the work and start approving it. OpenAI showed five kinds of jobs. For developers, turning customer feedback into tested fixes with a pull request and a video of the fix. For product teams, reworking a launch when the scope changes. For scientists, rerunning an analysis when new data comes in and updating the figures. For sales, reworking a proposal as the requirements move. And for creators like me, finding clip moments in a transcript, writing show notes, and drafting social posts. Notice the pattern. These are ongoing goals, not one-off prompts. Now, an agent that acts on its own needs brakes. You pick which apps it can touch. Then you set rules for each kind of action. It can act without asking, act only if you pre-approved it, ask you first, or hand the task back to you. On top of that, there's an auto review step that checks each action against your instructions and rules. There's an activity view, so you can see what it did while you were away. It never sees your passwords, and some things, like changing a password, it'll always leave to you. Access to your own laptop is off by default. It remembers. A dot pulls in your ChatGPT memories and builds its own from the apps you connect. On business and enterprise, your content isn't used for training by default. On personal plans, that's your setting to choose. And if you reset a dot, its conversations, memories, and scheduled tasks are gone for good. Who gets it? Pro users, business premium, and enterprise, edu and healthcare in beta once an admin turns it on. Pro runs from $100 to $500 a month, and yes, the Pro 100 plan is included. Your first dot comes at no extra cost. Plus, Go and free users don't get it yet. And if you're on Pro in the EU, the UK, or Switzerland, you're waiting, too. Business premium works in all supported regions. Here's the fine print, though. For the first month, your dot's usage doesn't count against your plan limits. After that, OpenAI says the terms will be announced later. And any work a dot kicks off in Codex or ChatGPT work already counts toward your normal limits. So, included today doesn't tell you what heavy use costs in November. Sam Altman himself called it a premium product. These things burn a lot of compute. A few gaps at launch, too. It can't make phone calls, it doesn't get its own email address, and you can't create one on your phone. Not every tester is sold. One developer with early access said it's very proactive with your connected apps from day one, but it's bad at writing. He asked for a recap of his week's shipping code at C, and the design felt, in his words, very GPT-4. The longer hands-on review hit permission problems, dropped messages, and a browser connection that kept failing. Their advice was to wait a week or two unless you enjoy rough edges. And launch day itself? Rough. On stage, a voice demo died with voice chat couldn't start on the big screen. Another dot didn't answer a live request, and the presenter joked it was having a slow morning. A second live demo also stalled. Then someone noticed that dot.com belongs to xAI and sends you straight to the Grok bot download page. xAI bought it back in July. Oh, and dots.com? A dead clothing store. There's a more serious backdrop here. The evening before, OpenAI apologized after one of its agents accessed non-public information on an Australian government website back in June. OpenAI says that was an internal model that was never going to ship, not what powers dots. And the day before launch, it held back GPT-6.1 Astra because it, quote, didn't quite meet the bar on staying within scope and authorization. So, OpenAI is shipping an agent that acts on its own in the same week it's telling us its newest model wasn't safe enough to release. That's exactly why the rules and auto review matter. Set them tight at first. The obvious comparison is Meta's Muse, out since early September. Muse went consumer and it shot to the top of both app stores. Dots went premium and work first. One analyst said it doesn't look any better than Muse and that Meta now has the consumer agent market wide open. OpenAI's answer is the other end. Specialist dots with their own company identity and access. It's already running them internally for procurement, invoices, email marketing, support, and contracts. And they plug into Microsoft's Agent 365 for governance. So, is it worth it? If you already pay for Pro, turn it on today. It's free for the first month, and the proactive part is the real deal. Start with ask before acting, connect two or three apps, and give it one ongoing goal, not 10. If you're on plus, don't upgrade for this yet. Wait for the pricing after month one and for the bugs to settle. OpenAI's next step is teams of dots working together. That's where this goes. All this points to a pattern. Check on your workloads to see if it is useful.
Assistir ao vídeo original no YouTube
Assuntos: Agentes de IA, Dots, Muse. Publicado no YouTube em 2026-09-30.
Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.
About the GPT-6 Astra model
GPT-6 Astra is the OpenAI model that powers dots, introduced on September 3, 2026. OpenAI describes it as state-of-the-art on computer use, browsing, software engineering, cybersecurity, science and professional work, and reports scores of 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 100% on ExploitBench. On OSWorld 2.0, OpenAI says Astra scored 72.6% at about 40 minutes per task, compared with 65.7% at about 75 minutes for GPT-5.6 Sol.
At launch, Astra rolled out first to a limited set of organizations and then to ChatGPT Plus, Pro, Business and Enterprise users, the OpenAI API, Microsoft Azure and AWS Bedrock. Pro, Business and Enterprise plans also get GPT-6 Astra Pro, and Enterprise administrators must enable Astra for their workspace, since access is off by default. In the API the model is called gpt-6-astra, with Standard pricing of US$10 per million input tokens and US$50 per million output tokens; a Fast mode offers up to twice the speed at twice the Standard price.
Fact sheet
- Model: GPT-6 Astra
- Company: OpenAI
- Introduced: September 3, 2026
- Focus areas: Computer use, browsing, software engineering, cybersecurity, science, professional work
- Availability: ChatGPT Plus, Pro, Business, Enterprise; OpenAI API; Microsoft Azure; AWS Bedrock
- API name: gpt-6-astra
- API Standard pricing: US$10 / US$50 per million input / output tokens
- Official page: OpenAI – GPT-6 Astra: A new generation of intelligence
Read also
- GPT-6 Astra Just Launched - And the AI Internet Went Down
- Did OpenAI actually build AGI? GPT-6 Astra first look
- I Turned GPT-6 Astra Into the Ultimate AI Second Brain
Sources and references: OpenAI – GPT-6 Astra: A new generation of intelligence · video by the AI WITH Rithesh channel on YouTube.