OpenAI DevDay: dots, GPT-6.1 Sol, the Pro Pricing Shuffle, and a Few Days With Muse
OpenAI's DevDay was all about dots, ChatGPT's new always-on agents — a catch-up flex on Meta's Muse — plus GPT-6.1 Sol, an Ultrafast tier, and a Pro pricing reshuffle that won't please everyone. My dot hasn't shown up yet, so the hands-on half is a few days with Muse, where the trust question gets awkward.
OpenAI’s DevDay just wrapped, and persistent agents were the main theme — a catch-up flex on the other always-on, always-working systems out there, Meta’s Muse chief among them. The big question: did OpenAI cook, and if so, how much is that meal going to cost? The full rundown’s in the video.
One caveat: my dot hasn’t shown up yet — I’m on Pro, and I’m always the last one — so everything about dots below comes from the keynote and OpenAI’s docs. The hands-on half is Meta’s Muse.
What dots are: agents that keep working when you don’t
The leaks called it Aeon, then “O,” before OpenAI finally revealed dots. Not to be confused with the Halloween candy, which is not good candy.
The pitch is agents that work on your behalf 24/7, whether you’re at the computer or not and whether Codex is open or not. Each dot gets its own little mini computer — OpenAI calls it a cloud computer — somewhere on OpenAI’s servers, plus its own browser where it can write and test code. I don’t know whether that’s Linux or a stack of Chromebooks; what matters is that each one is sandboxed from everyone else’s. It runs on GPT-6 Astra (more in my AGI-claim video). You message it or call it in ChatGPT, or reach it through Slack or Teams, and it can contact you the same ways.
Sound familiar? OpenClaw, Muse, and “trivial”
If this sounds familiar, it is a bit of a dressed-up version of OpenClaw. Its creator, Peter Steinberger, joined OpenAI back in February, and OpenClaw is now one of the partner apps.
The other comparison is Muse, which despite anyone’s reservations about Meta and privacy has been quite a hit — it knocked ChatGPT off the number one spot in the US App Store. (If you only know Muse as an image model, that’s where it started for me.)
OpenAI is aiming at more ambitious work than “book me a flight” or “order me those refrigerator filters I know you haven’t changed in three years.” In a pretty brutal diss, Sam called that kind of usage trivial. The pitch is that you hand a dot a boatload of responsibilities the way you’d hand them to a production manager, and it keeps chewing away behind the scenes. Great, and still not doing your laundry or mowing your lawn. We’ve got to wait for the robots for that.
ChatGPT Space and the magic data prisons
Space is basically OpenAI’s Google Drive, with agentic pages where you, your teammates, and your dot collaborate. I get what they’re doing — a walled garden around your information — and I’m not 100% sold. No shade; they’re all doing it. It’s an interesting carrot dangle to see how they herd everybody into their little magic data prisons.
The “oh wow” call, and one step closer to Jarvis
The demo walked a fictional music platform through development, design, and release, with the planned “oh wow” moment being a voice call to the dot. That call was slow, and it wouldn’t surprise anyone who’s been using voice lately.
Still, I’m being pragmatic, and I do think dots are pretty awesome. It legitimately brings us closer to — well, you’ve seen the rooms drenched in LED lights and “Jarvis, make me $2 million in the next 24 hours, make no mistakes.” Shout-out to Eric Morrison’s takedown of those clowns, AI LARP Cringe Lords. They’re about to go ham with dots.
My reductive take: it’s a little computer on OpenAI’s servers with hooks and cron jobs running. This is not your conscious AI agent. These are bots with timers and jobs that can cross-reference one another. Talking to a dot apparently doesn’t eat into your usage, though work it kicks off in Codex or ChatGPT Work does.
The catch: who gets a dot
The shoe drop: you’ll only find dots on Pro, Business Premium, and Enterprise, in beta starting the day of the keynote, rolling out gradually. Pro users in the EEA, UK, and Switzerland are excluded for now.
One correction to the video: I call the $100 Pro tier new and say dots cost $200. Neither is right. The $100 Pro tier has been around since April, and dots start there.
| Plan | dots? | Notes |
|---|---|---|
| Free, Go, Plus | No | |
| Pro, from $100/mo | Yes | Entry point |
| Pro, $200/mo | Yes | Reopened, smaller allowance |
| Pro 500, $500/mo | Yes | New; 25x Plus usage |
| Business Premium | Yes | |
| Enterprise | Admin-enabled beta |
No GPT-6.1 Astra, but GPT-6.1 Sol and Ultrafast
The teased GPT-6.1 Astra didn’t show up. It was reportedly shelved after safety testing, per Engadget. The reason I give in the video was a joke. To keep it straight: dots run on GPT-6 Astra; the shelved one is 6.1.
What we got is GPT-6.1 Sol: a fifth of Astra’s token price and, per OpenAI, near-Astra intelligence. A fix to the video: I say it’s available to everyone, but per OpenAI it’s in the API plus ChatGPT Work and Codex for Plus and up, not yet in regular Chat. There’s also Ultrafast, a speed tier of up to 300 tokens a second, in the API and on the new $500 plan.
The pricing no one liked
That new plan is Pro 500: $500 a month, 25 times the usage of Plus. They’re also bringing back the $200 Pro tier, with about half the usage it used to have. Per Engadget, Codex goes from 20x to 10x Plus and Pro model messages from 200 to 100 a week, and existing subscribers reportedly drop to the smaller allowance after October 29 at the same price. The tier breakdown is on OpenAI’s help page.
OpenAI’s reasoning is that the models are more efficient now, so you get more work done for less. We’ll see about that. In their defense, Opus 5.5 seems extremely token-efficient, so maybe they’ve figured something out.
Decisions API: basically it’s Jev
The one that made me chuckle: a previewed Decisions API, a super-fast model that picks from a short list of options — yes, no, send this here or there. Basically, it’s Jev. The New Stack framed it as OpenAI’s answer to Jev. Fair game; I never found a reason to play with Jev.
A few days with Muse
I went a bit YOLO, Claude’s finger-wagging notwithstanding, and gave Muse my inbox and calendar. I have to give Meta credit: it’s a pretty good product. I had it book a DMV appointment to renew my license, and it worked. I still ended up in line for 30 minutes, but that wasn’t Muse’s fault. Its news feed, after a day of tuning, is pretty good too.
I swapped Jolly the fuzzy marshmallow for Flamethrower Girl, then had it research her personality from my videos. Four days of hammering, and I’ve used about 50% of the free plan. It’s running Spark, which puts its intelligence somewhere around Opus level. But I don’t think it’s meant for power users — it’s more a product for soccer moms, while dots are clearly aimed at heavier lifting.
The trust question
Some of my frustration comes down to it being Meta. I remember Cambridge Analytica, and I’m sure you do too. That’s somewhat hypocritical of me, given OpenAI and Anthropic pretty much have free rein on my computer. But Meta has a rep, and it’s on them to correct it. So it feels a little sus when I want to give Muse sandbox access to one folder on my desktop and find out it needs my full computer. That’s not earning it back.
DevDay looked like an open shot OpenAI had to take: a preview of OpenAI Private Intelligence, a business feature where your data isn’t kept. They didn’t name Meta. I’m not sure where the line sits, though: without Private Intelligence, which I presume is paid, can OpenAI train on or look at your data? I don’t know.
Muse gets connectors, and Paperclip Heart
My other big frustration with Muse was its lack of connectivity, and that got fixed the day of DevDay, when Meta announced a whole boatload of connections. So it’s OpenAI’s serious dot, powered by Astra and priced accordingly, against Muse, more built for trivial things but with better integrations — and it’s Meta. All of these companies are fighting to become your new AI executive assistant and best friend, and it’s a battleground to watch.
Digesting everything from Meta Connect, especially the VR goggles and the glasses, it reminded me of the AI film I made a while back, Paperclip Heart. The YouTube algorithm kind of buried it, so not everyone saw it, but it’s relevant here.