TL;DR: OpenAI launched Dots at DevDay on September 29, 2026 — always-on GPT-6 Astra agents that keep working after your chat ends, each on its own cloud computer and browser. The first dot is included with ChatGPT Pro (from $100/month) and Business Premium; there is no API and no cheaper path. The same keynote halved Pro 200’s usage allowance and added a $500 Pro 500 tier. For most developers, a scheduled Claude Code or Codex job does 80% of this for $0 extra.

	OpenAI Dots	Cursor background agents	Claude Code + cron
Best for	Cross-app chores that outlive a session: inbox triage, monitoring, research, delegated Codex tasks	Repo-scoped coding tasks you launch and review the same day	Recurring repo jobs on a schedule you fully control
Price / cost	First dot inside Pro ($100/mo) or Business Premium; extra dots unpriced	Cursor plans from $20/mo + usage billing	The Claude plan or API key you already pay for; cron itself is $0
The catch	After the first month, dot-started Codex and Work tasks eat your plan allowance; eligible markets only	Stops when the task ends — nothing persists or watches overnight	You write, schedule, and babysit the automation yourself

Honest take: Skip Dots for now unless you’re already on Pro 100+ for other reasons — then claim your included dot and work it hard during the first-month free-usage window. Nobody should upgrade from Plus to a $100 plan just to get a background agent when a crontab line driving Claude Code or Codex CLI replicates the scheduled-coding use case for free.

OpenAI’s Dots announcement pulled roughly 750 points on Hacker News within two days of the September 29 DevDay keynote, and the pitch lands directly on territory that Cursor’s background agents and Claude Code’s scheduled workflows have been fighting over all year: an agent that does useful work while you’re not at the keyboard. The difference is architectural — and so is the bill. Here’s what’s confirmed, what’s still unpriced, and the arithmetic for deciding whether any of it belongs in your stack.

What is OpenAI Dots and how does it actually run?

A dot is a persistent agent inside ChatGPT, powered by GPT-6 Astra, that runs on its own OpenAI-hosted cloud computer with its own browser and keeps working toward a standing goal between your conversations. You talk to it by messaging or calling it in the ChatGPT app, or by pinging it from Slack or Microsoft Teams. It can monitor things, do multi-step research, draft documents, operate apps, and bring finished work back for review — and per multiple launch reports it does this proactively, without waiting for your next prompt.

Three structural facts matter more than the demo reel:

• It is cloud-only. The dot’s computer is OpenAI’s, not yours. There is no local runtime, no desktop binary, and no self-hosted option (more on the fake “download” below).
• It rides your existing connector grants. Dots connects through the same app permissions you’ve granted in ChatGPT, ChatGPT Work, and Codex — a grant in one applies across all four — reaching an ecosystem OpenAI counts at over 4,000 apps.
• There is no Dots API. As of October 3, 2026, you cannot trigger a dot programmatically from your own tooling or wire one into a CI pipeline. It is a consumer-surface product living in the ChatGPT app.

That last point answers the question most readers of this site will ask first: Dots is not a BYOK-able backend, not a Cursor competitor at the editor level, and not something Cline can call. It sits a layer above your coding tools and, at most, delegates to one of them (Codex).

What can a dot do for a coding workflow today?

The confirmed coding path runs through Codex: a dot can start tasks in Codex, and those tasks count against your plan’s Codex usage. In practice that means the overnight loop looks like: you give the dot a standing goal (“keep an eye on new issues in this repo, investigate reproducible bugs, prepare fixes”), the dot watches whatever surfaces it’s connected to, and the actual code work happens in Codex cloud tasks that hand changes back as pull requests to review — the same Codex mechanics we covered in the Codex CLI review.

The surrounding DevDay releases fill out the picture for developers:

• A new Code Review experience in the ChatGPT desktop app pulls PR summaries, changed files, comments, and checks into one interface; GitHub review is generally available, GitLab support is in preview (per the DevDay announcements reported September 29).
• @ChatGPT for Slack and Microsoft Teams shipped the same day, which is also how you ping a dot from chat.

What is not confirmed: that a dot itself writes or pushes code from its own cloud computer, outside Codex. Launch coverage describes the dot’s own machine doing research, browsing, and document work; code lands through Codex delegation. If your mental model is “a junior dev with commit rights running 24/7,” that is not what shipped — and given the permission model below, OpenAI clearly doesn’t want it to be.

How does the Dots permission model work?

Every supported action falls into one of four behaviors you set per rule: act without asking, act if pre-approved (meaning you explicitly requested that action in your prompt), ask before acting, or hand off to you entirely. On top of the configurable rules sit hard edges OpenAI says are enforced in code, not in the prompt:

• Background research tasks are read-only — they can read permitted sources and save private notes, but cannot send messages, change content in connected apps, or control a browser or computer.
• Password changes and money transfers always hand back to you, with no setting to override.
• Permanently deleting data and installing software can be gated to per-instance approval every single time.

This is a meaningfully more conservative default posture than most coding agents ship with — Cline asks before shell commands, but the “always comes back to you” category here is not user-disableable at all. For anyone who read our coverage of the MemGhost agent-hijack threat model, the structural worry with an always-on agent is obvious: it holds live grants to 4,000-odd app connectors around the clock, so the blast radius of a prompt injection in anything it reads is the union of everything you’ve connected. The code-enforced read-only research tier is OpenAI’s answer to exactly that. It’s the right design; it also means the autonomous mode that makes Dots interesting is deliberately fenced away from the risky actions that would make it maximally useful.

What does Dots cost — and what did DevDay quietly do to Pro plans?

Your first dot is included with ChatGPT Pro and Business Premium; Enterprise, Edu, and Healthcare workspaces get an admin-enabled beta. Additional dots and higher per-dot workloads are coming “later” with no announced pricing — Sam Altman called Dots “a premium product” that “uses a lot of compute.” Two usage rules decide what it really costs:

1. Conversations with a dot don’t count against your ChatGPT usage limits.
2. Tasks a dot starts in Codex or ChatGPT Work do count against those allowances — except during the launch promo, which exempts dots usage from eligible Pro, Business, and Enterprise allowances for the first month.

Which makes the same-day plan restructuring the fine print that matters. Verified against The Next Web’s and other DevDay reporting from September 29–30, 2026:

Tier	Price	Usage vs Plus	Changed at DevDay
Pro 100	$100/mo	5x	Includes your first dot
Pro 200	$200/mo	10x (was 20x)	Allowance halved — immediately for new subscribers, October 30 for existing ones; sign-ups (paused since Sep 10) reopened
Pro 500	$500/mo	25x	New tier; only plan with Ultrafast, serving GPT-6 Astra tokens up to 8x faster in Codex

Read those two tables together and the launch promo looks different: dots drive Codex usage, Codex usage bills against allowances that just shrank on the mid tier, and the only tier that got more headroom costs $500. A Pro 200 subscriber who lets a dot run nightly Codex tasks will discover in November that the same workload consumes twice the share of a halved allowance. That’s not a conspiracy, it’s a funnel — but you should see it before you aim your dot at a busy repo.

Dots vs Cursor background agents vs a scheduled Claude Code loop

For the specific job of “coding work happens while I sleep,” the three approaches price out very differently. Cursor’s background agents (covered in our Cursor 3.0 review) run repo-scoped cloud tasks from a $20/month plan with usage billing. Claude Code does the same from a terminal, and its scheduled-automation story — covered in the Dynamic Workflows guide — needs nothing fancier than cron:

# crontab -e — nightly issue triage at 02:00, no new subscription required
0 2 * * * cd ~/code/myrepo && claude -p "Triage issues labeled 'bug' updated yesterday; reproduce what you can; write findings to TRIAGE.md" >> ~/logs/triage.log 2>&1

Expected result: a dated TRIAGE.md in the repo every morning and a log line per run — the unglamorous version of “an agent worked while you slept,” at zero incremental subscription cost.

The honest decision math for a solo developer already paying $20/month for ChatGPT Plus:

• Upgrade to Pro 100 for the dot: +$80/month, $960/year. You get genuine cross-app persistence (inbox, Slack, browser-based monitoring) that neither Cursor nor a cron job replicates, plus 5x Plus usage.
• Stay put, schedule your existing tools: $0 extra. You get the repo-scoped 80% of the value — scheduled reviews, triage, test runs, dependency checks — with none of the cross-app reach.
• Already on Pro 100/200 for usage reasons: the dot is free marginal capability. Claim it, point it at low-risk standing goals, and burn the first-month exemption while it lasts.

The cross-app reach is the entire value gap. If your “while I sleep” wishlist is purely code-shaped, $960/year buys you nothing a crontab line doesn’t. If it includes “watch these three dashboards, summarize the on-call channel, and queue Codex fixes for anything that looks like our bug” — that’s the thing only Dots currently does in one product.

Don’t install “OpenAI Dots” from GitHub — it doesn’t exist as a download

Within days of the keynote, a GitHub organization named OpenAI-Dots appeared, offering a repo of the same name with a “Windows 10/11 zip, extract and run — official free download” pitch, surfacing in ordinary search results for Dots. Everything about the product’s confirmed architecture says this is fake: Dots is cloud-only, lives inside the ChatGPT app, has no desktop binary, and OpenAI’s real repos live under github.com/openai. A zip-with-executable riding a launch news cycle is the standard malware distribution pattern we’ve documented across the Open VSX evil-twin extensions and Amazon Q credential-theft incidents. If you want Dots, the only doors are the ChatGPT app on a Pro or Business Premium plan. If a teammate already ran such a zip, treat it as a credential-theft incident: rotate GitHub and OpenAI tokens and audit OAuth grants.

Honest take: adopt now or wait?

Wait, unless the subscription math is already sunk. Concretely:

• Adopt now if you’re on Pro 100+ or Business Premium anyway: the first dot is included, the first month’s usage is exempt, and the read-only research tier plus approval gates make the downside manageable. Standing goals that play to its strengths: monitoring and summarizing, dependency and CVE watching, and queuing well-scoped Codex tasks for your morning review.
• Wait if you’d be upgrading specifically for Dots: additional-dot pricing is unannounced, there’s no API, rollout is limited to eligible markets, the Pro 200 halving shows allowances can move under your feet mid-subscription, and the post-promo cost of dot-driven Codex usage is exactly the number OpenAI hasn’t let anyone compute yet. Revisit when extra-dot pricing and a second month of real-world usage reports exist.
• Ignore it if your automation needs are repo-shaped: schedule Claude Code or Codex CLI with cron and spend the $960/year on something with a receipt. If what you actually want is autonomy with zero cloud dependency, the self-hosted route — an open-weight model behind Ollama driving a local agent, per runaihome.com’s local coding LLM guide and the FOSS agent landscape at aifoss.dev — remains the only version where the always-on computer is yours.

One prediction we’ll check back on: GPT-6.1 Sol launched the same day at one-fifth of Astra’s price, and nothing stops OpenAI from eventually running cheaper dots on Sol-class models. If extra dots arrive priced like Sol rather than like Astra, the math in this article changes — that, not the launch keynote, is the moment Dots becomes a mass developer product.

FAQ

Does OpenAI Dots work with Cursor, Cline, or other coding tools? No. There is no Dots API or integration surface as of October 3, 2026. The only confirmed coding delegation path is Codex: a dot can start Codex tasks, which bill against your plan’s Codex allowance.

Which ChatGPT plan do I need for Dots? Pro (from $100/month) or Business Premium includes your first dot; Enterprise, Edu, and Healthcare get an admin-enabled beta. Plus ($20), Go, and free tiers get nothing. Additional dots will cost extra, with pricing unannounced.

Do dot conversations burn my usage limits? Talking to a dot doesn’t count against ChatGPT limits. Tasks the dot launches in Codex or ChatGPT Work do count — exempt only during the first month after the September 29, 2026 launch.

Can a dot take destructive actions on its own? OpenAI’s documented rules put every supported action into one of four behaviors (act freely / act if pre-approved / ask first / hand off), with code-enforced floors: background research is read-only, and password changes or money transfers always return to you.

Is there a Dots download for Windows or Mac? No. Dots runs on OpenAI’s cloud infrastructure inside the ChatGPT apps. Any “OpenAI Dots download” zip — including the OpenAI-Dots GitHub organization circulating since launch week — is not from OpenAI and should be treated as malware.

Sources

• OpenAI DevDay 2026 live updates — CNBC, Sep 29 2026
• OpenAI launches Dots, its bubbly agentic avatar — TechCrunch, Sep 29 2026
• OpenAI halves Pro 200 usage and launches a $500 ChatGPT plan at DevDay — The Next Web
• OpenAI dots, explained from the docs: the permission model of an always-on agent — DEV Community
• OpenAI Dots pricing 2026: Pro, Business Premium, and real costs — eesel
• OpenAI unveils Dots, always-on ChatGPT agents at DevDay — AI Weekly
• OpenAI Launches AI Agent ‘Dots’ — Yahoo Finance
• OpenAI launches dots, always-on ChatGPT agents with their own computers — BetaNews

Transparency note: openai.com and help.openai.com were not directly reachable from this environment at verification time; every claim above was cross-checked against at least two independent reports of the official announcement and documentation, dated September 29 – October 2, 2026.

Last updated October 3, 2026. Pricing, plan allowances, and rollout status change frequently — verify against the official OpenAI pricing page before subscribing.

Know which coding tool is worth paying for

Hands-on comparisons of AI coding assistants and what each one costs to run — including the local-model path. Sent only when something changes. Unsubscribe anytime.

Email address Subscribe