AI News, Sep 29 (Evening): OpenAI's Dots and a Halved Pro
OpenAI used its DevDay keynote to sell an agent that works around the clock, and used the same event to cut in half what its $200 plan buys. This update covers Tuesday from 9 a.m. to 5 p.m. Eastern.
The Big Story: OpenAI’s Always-On Agent Starts at $100 a Month
OpenAI launched dots at DevDay in San Francisco. They are always-on agents running on GPT-6 Astra, each with its own cloud computer and browser, reaching more than 4,000 apps through plugins. You message or call your dot in ChatGPT, or message it in Slack and Teams. Sam Altman pitched it on stage as something you “delegate ambitious pieces of work” to, “the way you would to a high agency engineer or a chief of staff.”
Nobody gets one cheaply. Dots come only with ChatGPT Pro, which starts at $100 a month, or a Business Premium seat. Free, Go and Plus are left out, and so, for now, are Pro subscribers in the EEA, Switzerland and the UK. OpenAI’s help center says dots usage won’t count toward plan allowances “for the next month,” after which “we’ll share usage terms for each plan.” The real cost of running one hasn’t been published yet.
The fine print is narrower than “built to handle everything.” When you aren’t talking to it, a dot does “proactive research” through connected apps restricted to read-only, and it runs the reminders and recurring checks you schedule. Nothing in OpenAI’s docs describes a dot waking up because an email arrived. Purchases need your approval, and changing a password or moving money gets handed back to you. It also can’t have an inbox of its own: at launch “you cannot give your dot its own standalone email address,” so it works through your personal account. That is a step back from the leak. On Saturday TestingCatalog found a “-o” email suffix configured for the agent, then codenamed “o.”
TechCrunch noted that “much of this functionality was already possible through Codex and similar agentic harnesses.” Axios framed dots as OpenAI’s answer to Meta’s Muse. Hours before the keynote, Meta launched Muse for Small Business with connectors for QuickBooks, Shopify, Slack and Stripe, and Muse is still free for most uses. The Hacker News thread (385 points, 299 comments) spotted the gap: “Free, Go, and Plus are probably the most popular consumer-facing plans, and Dots isn’t available on any of those.”
On Monday OpenAI cancelled GPT-6.1 Astra because it didn’t stay “within scope and authorization.” A day later it shipped its most autonomous product on GPT-6 Astra, the model that one was meant to succeed. Our dots explainer has the setup details.
$500 buys the old deal; $200 now buys half of it
The same keynote introduced Pro 500, a $500-a-month plan with OpenAI’s highest usage allowance, which OpenAI put at “25x Plus.” It is the only Pro plan with Ultrafast, which OpenAI says runs up to eight times faster than standard in Codex and, per its Codex docs, burns included usage at eight times the standard rate.
Pro 200 is open to new subscribers again, with a smaller allowance. From October 30, its included usage in ChatGPT Work and Codex falls from 20 times the Plus allowance to 10 times, and GPT-6 Pro messages in chat drop from 200 a week to 100. The price stays $200. OpenAI’s email to existing subscribers says they keep current limits through October 29 and get a one-time grant of 62,500 usage credits, valued at $2,500 and expiring December 31. Codex engineering lead Thibault Sottiaux previewed the cut on X the night before: “it will net out at half the dollar in API spend compared to the old Pro $200 plan.”
OpenAI’s help page says the lower allowance reflects “our increasingly efficient models.” One reply on the Pro 500 thread (157 points, 163 comments) quoted that line and added “actual lol.” On OpenAI’s own developer forum, a Pro 200 subscriber wrote: “I’m asking them not to substantially reduce what we are already paying for.”
CFO Sarah Friar gave the rationale to CNBC: “When we launched our $200 SKU, people thought we’d lost our minds,” and customers are now “moving more towards consumption.” The $100 agent and the halved $200 plan are the same move. OpenAI is pricing by work done, not by seat, and dots’ own usage terms are due in a month.
Today’s Top Stories
GPT-6.1 Sol: near-Astra performance at a fifth of the price
GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens in the API, with cached input at $0.10, half of GPT-6 Sol’s cached price. It’s in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, but not yet in regular ChatGPT chat. OpenAI claims it beats Opus 5.5 by 2.2 points on AutomationBench at about a third of the cost. In one safety test it tried to get around explicit restrictions, like access-denied messages, in 23.5% of cases, against 64.4% for GPT-6 Sol and 17.4% for Astra. Sol’s Hacker News thread was the day’s largest at 639 points and 571 comments, and one of its most-answered comments summed up the day: “Ominous for the industry and investors that token price is becoming the main battleground.”
ChatGPT automations can now start from any plugin’s events
OpenAI’s release notes say “Supported plugins can use MCP events to start an automation when something changes in a connected app.” TNW calls it the proposed MCP Events standard. OpenAI’s help docs for event-triggered tasks in ChatGPT Work still list only three examples: new Gmail messages, new Slack channel messages and GitHub pull request activity. Business and Enterprise teams also get Team Tasks, which run on a schedule or when an event happens. With a protocol behind them, triggers can now come from any plugin that adopts it, though each app still has to ship support.
Anthropic says an open-weight model now matches Mythos at building exploits
Anthropic published an analysis of GLM-5.3, the latest model from Zhipu AI (Z.ai). On ExploitBench it built working end-to-end exploits in 50 of 410 attempts, against 56 of 410 for Claude Mythos Preview. Its safeguards gave way 64% of the time with a false cover story, 92% with prefilled reasoning and 100% once abliterated, while Claude models stayed at zero. NIST’s CAISI had already called it “the most cyber-capable open-weight model released to date.” Anthropic’s conclusion: “a critical threshold in freely accessible capabilities has now been crossed.” Project Glasswing was meant to give defenders a head start. By Anthropic’s own count, it lasted about five months.
Quick Hits
- More DevDay: plugin extensions put apps in the ChatGPT sidebar and in file viewers, with Canva, Figma and Adobe as early examples. Sign in with ChatGPT lets Plus and Pro users spend their plan’s usage in 16 partner tools, including Devin and Vercel, with a weekly cap per app. @ChatGPT now works in Slack and Teams channels for teammates without a ChatGPT licence. TNW has the full list.
- Email agents: A dot can’t have its own email address at launch; Carly gives each of its AI agents one you can cc, set to send on its own or draft for approval.
- Meetings: a new Meetings plugin takes notes without a bot joining, in beta on the Mac app. OpenAI requires users to get everyone’s consent, but the app’s reminder is visible only to the user.
- OpenAI’s money: Friar confirmed to CNBC 70% quarter-over-quarter growth and an enterprise business that has doubled since July, while CNBC reported early talks on a raise of around $30 billion. On stage, Altman said it would be “ill-advised” to be a newly public company right now.
- Outage: Claude.ai, Claude Code, Cowork and the API threw elevated errors from 10 to 11 a.m. Eastern, and some messages sent during that hour may not have been saved.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
