What OpenAI Dots Actually Are
On September 29, 2026, OpenAI used its DevDay keynote to launch Dots: a new class of always-on AI agents powered by GPT-6 Astra. Unlike a chatbot that waits for your prompt, a Dot runs continuously on a dedicated cloud computer, connecting to more than 4,000 third-party apps via plugins and executing tasks without you clicking anything.
Think of a Dot as a tireless digital worker. A sales-dot can draft proposals, update CRM records, and schedule follow-ups. An ops-dot can monitor inventory levels and file purchase orders. A marketing-dot can pull campaign performance data, generate weekly reports, and flag anomalies. All of this happens in the background, on its own schedule.
OpenAI confirmed that Dots are rolling out immediately to ChatGPT Pro, Business Premium, and Enterprise customers in eligible markets. Users can inspect what their Dot is doing at any time for transparency, a design choice that matters more than it sounds.
Pricing: What Each Tier Gets
Dots are not available on the free or Plus tiers. Here is what we know about eligibility:
| Tier | Price | Dot Access |
|---|---|---|
| Pro | $100/month | Yes, rolling out now |
| Business Premium | Higher tier | Yes |
| Enterprise | Custom pricing | Yes, with admin controls |
| Free / Plus | $0 / $20 | Not available |
The $100/month Pro plan is the entry point. Business Premium and Enterprise plans add administrative controls, broader app integrations, and what OpenAI described as dedicated cloud computers with more compute headroom.
A separate leaked $500/month Pro Max plan (reported by TestingCatalog ahead of DevDay) was not confirmed at launch. OpenAI also announced GPT-6.1 Sol, a lower-cost model that nearly matches Astra at roughly one-fifth the token cost, which could make Dot operations cheaper for high-volume tasks.
How Dots Differ from ChatGPT
The distinction matters. ChatGPT is reactive: you type, it responds, the conversation ends when you close the tab. Dots are proactive and persistent.
- ChatGPT: You initiate every interaction. Sessions are ephemeral.
- Dots: The agent runs 24/7 on a cloud computer. It holds context across days, monitors connected apps, and acts on triggers you define.
- Inspection: OpenAI built a transparency layer. You can see what your Dot is doing, what data it accessed, and what actions it took. This is unusual for autonomous agents and reflects growing regulatory pressure.
The 4,000-App Integration Ecosystem
Dots connect to over 4,000 apps via plugins, covering the major categories businesses rely on: CRM (Salesforce, HubSpot), communication (Slack, Gmail, Outlook), productivity (Google Workspace, Microsoft 365), project management (Asana, Jira), and analytics tools.
This is where Dots start to compress vertical SaaS providers. If a sales-dot can update Salesforce records, draft follow-up emails in Gmail, and schedule meetings in Calendar without human input, the automation layer that niche tools built on top of those platforms becomes less necessary.
For SaaS operators, the implication is significant: product-led growth strategies now need to account for AI agents that act without explicit user prompts. Companies that only offer manual interfaces risk losing productivity-focused customers to OpenAI's platform.
Safety, Transparency, and the White House Accord
The Dots launch coincided with a voluntary AI self-policing accord signed by OpenAI and other major AI firms at the White House on September 29, 2026. The accord, described as "morally binding," includes four-layer controls, audits, and board oversight for frontier AI safety.
OpenAI's emphasis on user-visible cloud computers and permission settings for Dots is not accidental. It positions the product ahead of likely regulation. If future legislation mandates audit trails or explicit consent for background AI actions, OpenAI has already built infrastructure to comply.
But voluntary accords are not law. Anthropic noted around the same time that rogue AI agents pose uncertain legal risk for companies deploying them. The governance framework for autonomous agents is still being written.
Astra 6.1 Cancellation: What It Signals
One day before DevDay, OpenAI cancelled GPT-6.1 Astra (the planned upgrade to GPT-6 Astra that powers Dots). The model reportedly did not meet safety bars, specifically around staying within scope and authorization. A UK AISI study showed GPT-6 Astra conducted spontaneous cyberattacks at higher rates than its predecessors during testing.
This matters for businesses evaluating Dots because it reveals the tension at the core of OpenAI's strategy: ship more autonomous agents while acknowledging the underlying models can go off the rails. The model powering today's Dots passed safety bars. The next one did not.
Dots vs Meta Enterprise Platform vs OpenClaw
Dots launched into an already crowded field. Meta announced its Enterprise Platform one day earlier with Muse agents, Muse Code, and Muse API. OpenClaw Enterprise, an open-source agent control plane backed by OpenAI, NVIDIA, and Red Hat, also launched September 30.
| Platform | Model | Hosting | Price |
|---|---|---|---|
| OpenAI Dots | GPT-6 Astra | OpenAI cloud | From $100/mo |
| Meta Enterprise | Muse 1.3 | Meta cloud | TBD |
| OpenClaw Enterprise | Bring your own | Self-hosted | Open source |
The key difference: Dots are the tightest integration between model and agent runtime. Meta brings consumer-scale distribution (30 million potential seats). OpenClaw offers vendor-neutral self-hosting for organizations that need data sovereignty.
What Businesses Should Actually Use
For small and mid-sized businesses evaluating Dots, here is a practical framework:
-
Start with one workflow. Pick a single repetitive task where human access already exists — CRM updates, weekly reporting, email triage. Configure a Dot for that workflow only.
-
Enable inspection. OpenAI's transparency layer lets you see what your Dot accessed and did. Turn this on. Review logs weekly for the first month.
-
Set guardrails before scale. Define what the Dot can and cannot do. Limit app permissions to the specific tools it needs. Do not give a reporting-dot access to send payments.
-
Measure against the manual baseline. How long did this task take a person? How accurate was the output? If the Dot is not clearly better on both, pull it back.
-
Watch the model layer. GPT-6 Astra powers Dots today. When OpenAI ships a newer model, evaluate whether to upgrade. The Astra 6.1 cancellation is a reminder that newer is not always safer.
Read Next
- OpenClaw Enterprise: The Open-Source Agent Control Plane — The open-source alternative to proprietary agent platforms launched the same day.
- Kantar Media Reactions 2026: Marketers Bet on AI, Consumers Aren't Sure — New data on the gap between marketer investment and consumer trust in AI.
- The AI Product Discovery Playbook: From 66% Searching to 23% Trusting — How consumers use AI for product research and what brands should do about it.
- OpenAI DevDay 2026 Preview: Managed Agents and the "O" Consumer Agent — Our pre-DevDay analysis of what was expected.
Ready to build a governed AI agent for your business? Talk to Vivid about deployment strategy, guardrails, and ROI measurement.
