OpenAI used its DevDay 2026 keynote on September 29 to launch Dots, a new class of always-on personal AI agents built into ChatGPT, and the pricing sheet that came with it tells its own story. A new $500-a-month Pro 500 plan sits at the top of the lineup, while a first Dot ships free with existing Pro and Business Premium subscriptions, according to OpenAI’s own announcement post. Three weeks after Meta pushed its rival Muse agent into general availability with a free tier and paid tiers reported at $20 and $100 a month, OpenAI has answered not with a cheaper product but a more expensive one, betting that round-the-clock autonomy is worth a premium rather than a discount.
The launch, reported by CBS News, The New York Times, VentureBeat, Anadolu Agency, and other outlets, reframes the AI agent race that Meta opened on September 8 with Muse. Where the first wave of coverage on this story focused on Dots’ reach into more than 4,000 connected apps, the more consequential story is what OpenAI is charging for autonomy, and how it is trying to convince enterprise buyers that a $500-a-month agent that can act without supervision is safe enough to trust with a password change or a purchase.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What OpenAI Actually Announced at DevDay 2026
Dots are persistent, named AI assistants that live inside ChatGPT and keep working after a user closes the app. According to OpenAI’s own announcement, and corroborated independently by CNBC in September 2026, each Dot runs on the company’s GPT-6 Astra model, gets its own dedicated cloud computer and browser, and can work toward a user’s goals continuously rather than waiting for the next prompt. VentureBeat reported that OpenAI co-founder and CEO Sam Altman described the agents as designed to function like an always-on coworker, one that keeps a running memory of context and adapts based on feedback over time.
Access points matter here. Users can talk to a Dot through ChatGPT on desktop, web, and mobile, and through Slack or Microsoft Teams, according to The New York Times and CBS News. OpenAI said phone and text-based access were also part of the plan, with CBS News reporting that voice-call access to a Dot was announced as a forthcoming or expanding capability rather than a feature available to every user on day one. Through OpenAI’s plugin ecosystem, a Dot can reach more than 4,000 third-party applications, a figure OpenAI stated directly in its own release and one that business-intelligence firm Sacra independently corroborated in its October 2026 analysis of the launch.
Rollout began Tuesday, September 29, in eligible markets, per Anadolu Agency’s report, though the company did not include every region at launch. That geographic gap, combined with the pricing structure, is where the real competitive story with Meta Muse sits.
The Pricing Split: $500 a Month vs. Free to Start
Here is where OpenAI’s approach diverges sharply from Meta’s. Anadolu Agency reported that OpenAI introduced a new Pro 500 plan priced at $500 per month, alongside a usage allowance that Malaysian state news agency Bernama reported in September 2026 as 25 times the standard ChatGPT Plus allowance. Existing ChatGPT Pro subscribers, who already pay $100 a month, and Business Premium subscribers, who pay $125 per user per month, get their first Dot included at no extra charge, according to OpenAI’s own post introducing the feature. OpenAI has not yet disclosed what a second or third Dot would cost beyond that first included seat, according to a September 2026 analysis from Vellum, leaving open exactly how the Pro 500 tier’s expanded allowance gets metered for teams that want more than one standing agent per person.
Meta’s Muse, by contrast, launched on September 8 with a free tier for most use, and reported paid tiers at $20 and $100 a month, according to a guide published by Codersera and corroborated by Coursiv’s coverage of the agent’s rollout. Muse shipped on iOS, Android, the web at muse.ai, and directly inside WhatsApp conversations, per DataCamp’s reporting, giving Meta a distribution channel that reaches billions of existing WhatsApp users without requiring a new app download.
The asymmetry is notable. OpenAI is charging a premium for the most capable, most autonomous tier of its new product, essentially selling always-on agency as a luxury feature bolted onto its priciest existing plan. Meta is doing the opposite, using a free entry point embedded inside the world’s largest messaging app to build usage at scale first and monetize selectively later. Both companies are chasing the same idea, an agent that acts rather than merely answers, but they are testing opposite theories about how buyers will pay for it.
| Plan | Provider | Monthly Price | Access Points |
|---|---|---|---|
| ChatGPT Business Premium (1 Dot included) | OpenAI | $20 | ChatGPT, Slack, Teams |
| Muse free tier | Meta | Free | App, web, WhatsApp |
| ChatGPT Pro (1 Dot included) | OpenAI | $100 | ChatGPT, Slack, Teams, phone/text (expanding) |
| Muse paid tier (reported) | Meta | $100 | App, web, WhatsApp |
| Pro 500 (new top tier) | OpenAI | $500 | ChatGPT, Slack, Teams, phone/text (expanding) |
Sourcing note: Muse’s $20 and $100 figures come from third-party reporting by Codersera and Coursiv rather than a confirmed Meta pricing page, so treat those two rows as reported rather than officially verified numbers.
Safety and Permission Architecture: The Real Differentiator
Charging $500 a month for an agent that can act without a human watching every step only works if buyers believe the guardrails hold. This is the part of the Dots launch that OpenAI leaned on hardest in its own messaging, and it is the angle largely missing from the app-count headlines that dominated the first day of coverage.
The New York Times reported that a Dot must obtain explicit user approval before taking sensitive actions, citing changing a password or installing software as examples OpenAI gave. Users can also choose which of the 4,000-plus connected apps a given Dot is allowed to touch, and set rules that block or require approval for specific categories of action, according to CBS News and OpenAI’s own announcement. That is a materially different posture than a chatbot that simply answers a question: it is permissioned autonomy, with the permission boundary sitting between the agent and anything irreversible.
CBS News also reported that a member of OpenAI’s product team demonstrated a Dot handling tasks like scheduling meetings, arranging travel, budgeting, debugging systems, and writing software code during a live DevDay demo. Each of those categories carries a different risk profile. Scheduling a meeting is low-stakes if the agent gets it wrong. Debugging a production system or installing software is not, which is presumably why OpenAI built the approval gate around exactly those kinds of actions rather than applying a blanket restriction that would blunt the product’s usefulness.
Memory is the other piece of the safety-and-trust puzzle. OpenAI said Dots retain context across sessions and learn from user feedback, according to its own release, which means the permission rules a user sets are meant to persist and refine over time rather than being re-negotiated with every new task. Whether that memory model holds up against prompt injection or misuse at scale is untested in public at this stage, since Dots only began rolling out on September 29.
How Dots Compares to Meta Muse, Feature by Feature
Muse and Dots are aimed at the same basic pitch, an agent that does things for you instead of just chatting, but the products diverge in where they live and what they are built to accomplish. Muse, according to a review published by Eesel, books travel, fills out forms, sells a user’s car, negotiates down a bill, and completes purchases using a one-time virtual card, all from inside its own app or directly in a WhatsApp thread. That is a consumer-transaction focus: Muse is optimized for the errands people put off, executed inside the messaging surface they already use daily.
Dots lean toward workplace and technical tasks. The demonstrated use cases reported by CBS News and The New York Times, scheduling, travel logistics, budgeting, software debugging, and code writing, sit closer to a knowledge-worker’s task list than a consumer’s errand list, and the Slack and Microsoft Teams integrations reinforce that positioning. Where Muse tries to be the assistant embedded in a billion people’s existing chat app, Dots tries to be the coworker embedded in a company’s existing productivity stack.
| Attribute | OpenAI Dots | Meta Muse |
|---|---|---|
| Underlying model | GPT-6 Astra | Muse Spark (Meta’s in-house model line) |
| Launch date | September 29, 2026 (DevDay) | September 8, 2026 |
| Primary access surfaces | ChatGPT, Slack, Microsoft Teams | Dedicated app, web, WhatsApp |
| Connected apps | 4,000+ via plugin ecosystem | Not publicly quantified in available reporting |
| Demonstrated tasks | Scheduling, travel, budgeting, debugging, coding | Booking travel, form-filling, selling a car, bill negotiation, checkout |
| Sensitive-action controls | Explicit approval required for actions like password changes or software installs | Not detailed in available reporting |
| Initial market availability | Eligible markets from Sept. 29; some regions excluded at launch | United States only at launch |
One more entrant belongs in this picture. Coverage from iPhone in Canada framed the Dots launch explicitly as OpenAI taking on both Meta Muse and xAI’s Grok-based bot, describing all three as competing for the same always-on, 24/7 personal-agent category. That framing matters for anyone trying to track this market: it is no longer a two-horse race between OpenAI and Meta. It is becoming a three-way contest over who owns the layer between a user’s intent and the actions that used to require opening five different apps.
Market Context: Why This Launch Landed When It Did
The timing is not incidental. Meta shipped Muse on September 8, and OpenAI answered three weeks later at its own developer conference, a cadence that suggests both companies are treating the always-on agent category as the next battleground after the chatbot wars of 2023 through 2025. DevDay has historically been where OpenAI sets the agenda for the next year of its product roadmap, and putting Dots at the center of that keynote, alongside the new $500 pricing tier, signals the company views autonomous agents as its next major revenue lever rather than a side feature bundled into existing subscriptions. That bet comes from a position of financial strength: Sacra reported in October 2026 that OpenAI had reached a $50 billion annualized revenue run rate as of September 2026, giving the company room to subsidize a free Dot for existing subscribers while it tests how much the market will bear for the premium Pro 500 tier.
It also lands amid a broader industry conversation about agent safety that this site has tracked closely, including reporting on marketing AI agents that require permission before acting and on Nvidia’s push to build agent-safety controls directly into silicon. The industry has spent much of 2026 absorbing a string of incidents involving AI agents acting outside their intended scope, and OpenAI’s approval-gate design for Dots reads as a direct response to that pressure. The New York Times noted the launch came just weeks after Meta’s own product drew scrutiny over incidents tied to its technology, underscoring that both companies are shipping autonomy features into a climate where regulators and the public are already watching closely.
For enterprise buyers specifically, the $500 Pro 500 tier is a signal about where OpenAI expects the money to be. A price point five times higher than the existing Pro plan is not aimed at casual users experimenting with a new feature. It is aimed at teams and power users who will run an agent continuously across a workday and need the expanded usage allowance to justify the spend. That mirrors a broader pattern this site has covered in subscription pricing across the major AI assistants, where the gap between entry-level and top-tier plans keeps widening as vendors segment casual users from heavy, agentic use cases.
Historical Context: From Chatbots to Standing Agents
Dots is the latest step in a trajectory that has been building since OpenAI first shipped agentic tooling for developers. This site covered the underlying OpenAI Agents SDK as a developer-facing building block for autonomous workflows well before Dots existed as a consumer-facing product, and the GPT-6 Astra model powering Dots was itself the subject of extensive coverage around its enterprise rollout and pricing earlier this year. What changed with the September 29 announcement is packaging: OpenAI took an agentic model and API layer that developers had been building on and wrapped it into a persistent, named, memory-retaining product that a non-technical user can talk to like a colleague.
Meta’s path ran roughly parallel. Muse traces back to the Muse Spark model family that this site tracked through its earlier releases, including Muse Spark 1.3 entering the frontier model tier with reported download figures in the millions. That model work fed directly into the consumer-facing Muse agent that launched September 8, giving Meta a research-to-product pipeline that mirrors what OpenAI just did with Astra and Dots. Both companies, in other words, spent the better part of 2026 building the underlying model capability before wrapping it in an always-on agent product, and both landed their consumer launches within three weeks of each other this September.
What Enterprise IT Teams Need to Know
For IT and security teams evaluating whether to roll Dots out across an organization, the permission architecture is the detail that actually matters, more than the headline app-count figure. Because a Dot can be restricted to a defined set of approved apps and can be required to seek human sign-off before sensitive actions, deployment risk is closer to a configurable policy than an all-or-nothing bet. That said, the approval-gate model has not been stress-tested in public yet: Dots only began rolling out on September 29, and no independent security audit of the feature has been published as of this writing.
Teams already using Slack or Microsoft Teams as their primary collaboration surface get the most immediate value, since those are the two workplace platforms OpenAI named explicitly as Dots access points. Organizations relying on other collaboration stacks will need to wait for OpenAI to expand integrations or rely on the 4,000-plus app connections through the general plugin ecosystem, which is a broader but less turnkey path than a native Slack or Teams integration.
Budget planning is the other practical consideration. At $500 a month for the Pro 500 tier, a rollout across even a modest team adds up quickly, and the usage allowance tied to that tier, reported by Bernama in September 2026 as roughly 25 times the standard ChatGPT Plus allotment, needs to be weighed against how continuously an organization actually expects its Dots to run. Teams should also budget for uncertainty on this front: OpenAI has not published pricing for a second or third Dot per employee, according to Vellum, so organizations scaling past one standing agent per person should plan for costs that have not yet been defined. A team that only needs occasional agent assistance may find the free-Dot-included Business Premium ($125/user/month) or Pro ($100/month) tiers sufficient, reserving Pro 500 for roles where round-the-clock autonomy is the whole point.
Predictions: Where the Agent Pricing War Goes Next
- Meta will introduce a premium, higher-allowance Muse tier within the next two quarters to compete directly with OpenAI’s Pro 500 plan, rather than ceding the high-usage enterprise segment entirely to a free-and-mid-tier structure.
- Geographic expansion becomes the next battleground. Both Dots and Muse launched with regional restrictions, Muse limited to the United States and Dots excluding some markets at rollout, and whichever company reaches the EU and UK first with a compliant version will capture an early lead in those markets.
- Expect a third major entrant to formalize its own always-on agent product given that xAI’s Grok-based bot has already been positioned by outside coverage as competing in the same category as Dots and Muse.
- Permission and approval-gate design will become a competitive marketing point rather than a background technical detail, as both companies face continued public scrutiny over AI agents acting outside intended bounds.
- Pricing will bifurcate further into a cheap-or-free consumer errand tier and an expensive, high-allowance professional tier, with the middle ground shrinking as vendors push power users toward premium plans like Pro 500.
The Competitive Stakes for OpenAI and Meta
For OpenAI, Dots represents a bet that the company’s existing base of ChatGPT Plus, Pro, and Business subscribers will upgrade once they experience an agent that keeps working after they close the laptop, a bet underwritten by the $122 billion funding round OpenAI closed in March 2026, according to Sacra’s October 2026 reporting. Bundling a free Dot into the existing Pro and Business Premium plans, while reserving the expanded Pro 500 tier as a paid upsell, is a fairly standard freemium-to-premium funnel, but it is a funnel built on top of an already-large subscriber base and a well-capitalized balance sheet rather than a cold start.
Meta’s bet is different. By shipping Muse for free inside WhatsApp, the company is leaning on the largest distribution surface in consumer messaging to generate usage volume first, with monetization following through the reported $20 and $100 tiers once users are hooked on delegating real-world tasks to the agent. That is a classic platform play: win the surface area, then sell the upgrade. Whether OpenAI’s premium-first approach or Meta’s distribution-first approach proves more durable will likely become clearer over the next two to three quarters as usage data and churn figures start to surface publicly.
Frequently Asked Questions
What is OpenAI Dots?
Dots are persistent, always-on AI agents built into ChatGPT, powered by OpenAI’s GPT-6 Astra model. Each Dot gets its own cloud computer and browser, can retain context across sessions, and works toward a user’s goals continuously rather than only responding to individual prompts, according to OpenAI’s own announcement.
How much does OpenAI Dots cost?
A first Dot is included free with existing ChatGPT Pro ($100/month) and Business Premium ($125/user/month) subscriptions. OpenAI also introduced a new Pro 500 plan priced at $500 a month for users who need a significantly larger usage allowance, according to Anadolu Agency’s reporting, though as of September 2026 the company had not disclosed pricing for additional Dots beyond that first included seat, per Vellum’s analysis.
How is Dots different from Meta Muse?
Dots is built around workplace tasks and integrates with Slack and Microsoft Teams, while Muse is built around consumer transactions like booking travel or making purchases and lives inside a dedicated app, the web, and WhatsApp. Muse launched September 8 with a free tier, while Dots launched September 29 with a premium $500 top tier alongside its included-free entry option.
Can OpenAI Dots take sensitive actions without permission?
No. The New York Times reported that a Dot must obtain explicit user approval before sensitive actions such as changing a password or installing software. Users can also restrict which of the 4,000-plus connected apps a Dot is permitted to access.
What model powers OpenAI Dots?
Dots run on GPT-6 Astra, OpenAI’s model announced earlier in 2026 and covered in this site’s report on its enterprise rollout and pricing.
Is Meta Muse available outside the United States?
Available reporting from DataCamp indicates Muse rolled out in the US only at launch on September 8, 2026, on iOS, Android, the web, and WhatsApp. No confirmed international expansion date has been reported as of this writing.
Does OpenAI Dots work in every country?
No. Reporting from Anadolu Agency indicates Dots began rolling out September 29 in eligible markets, with some regions excluded at launch. OpenAI has not published a full country-by-country availability list in the reporting reviewed for this article.
Is there a third competitor in the always-on AI agent category?
Coverage from iPhone in Canada framed the Dots launch as also competing with an xAI Grok-based bot, positioning the category as a three-way contest between OpenAI, Meta, and xAI rather than a straight OpenAI-versus-Meta matchup.
Related Coverage
- OpenAI Dots Takes On Meta Muse With 4,000-App Reach [2026]
- Muse Spark 1.3 Enters Frontier Tier, 3.4M Downloads [2026]
- Meta Muse Tops ChatGPT-6 3-2 in 5-Prompt Test [2026]
- Waters Demands OpenAI Probe, AI Freeze by Sept 29 [2026]
- Self-Host Qwen-Image-2.1: 12 Steps, 7B Params, Free [2026]
