OpenAI Dots
Persistent always-on agents inside ChatGPT, announced at DevDay on 29 September 2026; each Dot gets its own cloud computer, browser and memory, runs on GPT-6 Astra and keeps working on your goals between conversations.
Overview
One-line facts, source-labelled.
| Vendor | OpenAI |
|---|---|
| Form | Persistent always-on agent inside ChatGPT |
| Launched | September 29, 2026 (DevDay) |
| Model | GPT-6 Astra |
| ARscore | not scored |
| Pricing | first Dot included with an eligible paid plan |
| Successor to | ChatGPT Agent mode (discontinued) |
| Open source | No |
Model capability (DeepSWE 1.1 pass rate)
The OpenAI models behind the Dots line, charted by coding pass rate.
GPT-6 Astra
the model Dots run on
GPT-6.1 Sol
newer OpenAI flagship
Bar length equals the DeepSWE 1.1 pass rate. † marks a vendor self-claim that AgentRanks has not independently reproduced; unmarked bars come from the independent table. Source labels: Official vendor announcement, Independent AgentRanks measurement, Press third-party coverage.
Features
What ships, in OpenAI’s own terms (presence, not benchmark performance).
| Capability | Status | Source |
|---|---|---|
| Own cloud computer, browser and memory per Dot | ✓ | Official |
| Connects to 4,000+ apps via plugins | ✓ | Official |
| Reachable from ChatGPT desktop and mobile | ✓ | Official |
| Tag the Dot’s handle in Slack and Microsoft Teams | ✓ | Official |
| Voice-call access to the same Dot | ✓ | Official |
| Custom Rules: act / pre-approve / ask first / hand off | ✓ | Official |
| Activity View of delegated steps (desktop) | ✓ | Official |
| Texting a Dot | ◐ | Official (small US Pro beta) |
✓ supported · ◐ partial or beta · — not in scope. Presence, not benchmark performance.
Safety controls
The guardrails OpenAI ships with Dots.
| Background research | read-only by default; a watching Dot cannot send or change anything |
|---|---|
| Local computer access | off by default; enable per Dot only when needed |
| Hard-coded human gates | password changes and money movement always stay human |
| Auto-review | a separate system examines planned actions before they run |
| Prompt-injection claim | no scored successes in 16,600 generated attack emails (vendor) |
Setup and configuration requirements
Eligibility and first-run constraints as of October 2026.
| Eligible plans | ChatGPT Pro $100, $200 or $500 per month (adults; excluding the EEA, Switzerland and the UK); Business Premium $100 per user per month billed annually, $125 monthly, all supported regions |
|---|---|
| Enterprise, Edu, Healthcare | beta, off by default; a workspace admin must enable it |
| Not included | Free, Go ($8) and Plus ($20) plans at launch |
| Creation | desktop only: ChatGPT app (macOS, Windows) or chatgpt.com/dots; message it later from mobile |
| Rollout | gradual; may take several days even on an eligible plan |
| Usage limits | talking to a Dot is free; tasks it starts in Work or Codex count as usual |
Use cases
Examples OpenAI itself gives.
Living documents
Keep a sales proposal updated as requirements change.
Bug watch
Investigate a newly reported bug the moment it appears.
Content pipeline
Prepare drafts as new source material arrives, report back when ready.
Morning digest
A bounded first job: summarise new project emails every morning.
Download and official links
Outbound, for trial and verification.
| Product entry | https://chatgpt.com/dots |
|---|---|
| Announcements | https://openai.com |
| Desktop app | ChatGPT app for macOS and Windows, linked from chatgpt.com |
| News coverage | AgentRanks news feed |
Compare further
The other always-on bots and the scored terminal agents.