GPT-6 Astra Explained: What OpenAI's New Model Changes

OpenAI announced GPT-6 Astra at its Dev Day on September 29, 2026, as the model powering ChatGPT Dots, its new line of always-on agents. The announcement tells us a fair amount about what the agents do and much less about the model itself. This article separates the two, so you can see what is confirmed, what is still unpublished, and what it means for a business owner deciding whether to pay attention now or later.

Key takeaways

• GPT-6 Astra is the model behind ChatGPT Dots, described by OpenAI as remarkably capable, always-on agents built to handle ongoing work.
• Each Dot runs on its own cloud computer with a browser, connects to more than 4,000 apps, and can be reached through ChatGPT, mobile, Slack, and Microsoft Teams.
• Background research runs read-only by default, while actions can be allowed, blocked, or set to require approval, with sensitive tasks always staying with the user.
• Dots are rolling out to ChatGPT Pro and Business Premium in eligible markets, with an admin-enabled beta for Enterprise, education, and healthcare accounts.
• Benchmark scores, context-window figures, and API pricing for Astra itself were not published with the announcement, so treat performance claims as unverified for now.

What GPT-6 Astra is, based on what was actually announced

GPT-6 Astra is best understood as the engine inside a product announcement rather than a standalone model release with a specification sheet. OpenAI introduced it together with Dots, and the concrete details in the launch coverage describe agent behavior: persistent workers that keep going in the background, an Activity View where you can watch and redirect them, and proactive work such as research that runs without a prompt. Voice calls are supported, and SMS support is planned.

That framing matters because it tells you where the verified information sits. Everything about how a Dot behaves, where it runs, what it connects to, and how approvals work is documented across multiple launch reports. Almost nothing about the model weights, training, evaluation scores, or developer pricing was published alongside. If someone quotes you an Astra benchmark or a per-token price, ask for the source, because as of September 30, 2026, those numbers are not in the public launch record.

For a non-technical owner, the useful definition is short: Astra is the model that lets a Dot take a standing job, such as monitoring something or preparing something, and carry it forward without you re-prompting it each time. The ChatGPT Dots explainer covers the agent side in full, including channels and availability.

What Dots reveal about the model

Even without a specification sheet, the Dots design tells you what Astra was built to do. Each Dot gets an isolated virtual machine, meaning its own cloud computer and browser, rather than running on your laptop. It connects to more than 4,000 apps through OpenAI's plugin ecosystem, which positions ChatGPT as a place where work gets launched, not just a window where questions get answered. One analysis described this as a direct challenge to the app-store model, and that reading fits: if the agent can open the apps for you, the starting point of the work moves.

The approval design is the second clue. Background research runs read-only, actions can be allowed, blocked, or require approval, and sensitive tasks such as changing a password always stay with the user. Credentials can be used without exposing passwords to the model, and Dots can be stopped if systems detect malicious instructions. A model meant only for chat would not need that machinery. Astra is clearly meant to act inside real systems, which is why the control layer around it is as important as the model itself.

The third clue is reach. Dots work across interfaces and are reachable via ChatGPT, mobile apps, Slack, and Microsoft Teams, with voice calls supported now and SMS planned. Specialist Dots for procurement, invoice processing, customer support, and commercial contracts are in testing for businesses, with their own identities and access to selected company systems. OpenAI is working with Microsoft to bring those specialists under Microsoft Agent 365 security controls. That enterprise direction is worth noting even if you never buy it, because it signals where the governance expectations are heading.

What is still unknown, stated plainly

Three categories of information were not published with the announcement, and honest coverage should name them. First, benchmark scores: there is no verified public set of evaluation results for Astra itself in the launch record, so comparisons against other frontier models cannot be checked yet. Second, model access details: the announcement describes access through ChatGPT plans, not a developer API with published per-token pricing or rate limits. Third, the full regional picture: Pro is unavailable in the EEA, Switzerland, and the UK initially, while Business Premium is available in all supported regions, and Enterprise, education, and healthcare access depends on an administrator enabling the beta.

There is also a prior capability worth keeping separate. ChatGPT Work, an earlier long-running-jobs feature, counts its tasks toward the limits of the Codex or ChatGPT Work services that start them. Dots are a distinct product line with their own inclusion terms: the first Dot is included for Pro and Business Premium, with extended limits during the first month, and Dot usage does not count toward plan limits initially. Do not mix the two when you compare plans. Our Dev Day 2026 recap keeps the announced-versus-coming distinction in one place.

How a business owner should read launch claims

New-model announcements follow a pattern: the vendor emphasizes capability, the press repeats the strongest framing, and the numbers that would let you verify the claims arrive later. A practical filter has three steps. First, separate product facts from model claims. Product facts, such as the 4,000-plus app connections or the approval settings, are checkable in the launch coverage. Model claims without published evaluations are not checkable yet. Second, ask who the claim serves. An always-on agent that launches apps serves OpenAI's platform position as well as the user, which does not make it bad, but it does mean lock-in questions deserve attention. Third, translate the claim into a task you already pay for. If the agent cannot take a named weekly job off someone's plate, the model generation behind it does not matter to you yet.

A comparison table helps keep the categories straight:

If a vendor or a headline fills in the right-hand column for you, treat it as speculation until a primary source appears.

A sensible next move while details fill in

You do not need Astra's specification sheet to test whether delegated agent work helps your business. Pick one recurring job with a clear output, such as a weekly competitor watch, a lead-list cleanup, or an invoice-preparation pass, and run it as a 30-day pilot with approvals switched on. Define the inputs, the allowed actions, and who reviews the output before anything leaves the building. That pilot teaches you more than any benchmark, because it measures your task, your data, and your review cost.

Two guardrails apply from day one. Give the agent scoped credentials, never an owner's master login, and keep background work read-only until the approval behavior has proven itself on low-stakes tasks. The internal assistant service describes how scoped, supervised agents are typically set up for small teams, and the AI agent governance overview explains the control categories that larger buyers will ask about. If your pilot succeeds, expand to a second workflow; if it struggles, the failure will usually be in the brief or the permissions, not the model generation.

FAQ

What is GPT-6 Astra?

GPT-6 Astra is the OpenAI model that powers ChatGPT Dots, the always-on agents announced at OpenAI Dev Day on September 29, 2026. OpenAI describes Dots as remarkably capable agents that handle ongoing work, and Astra is the engine behind that behavior. Independent benchmarks and model-card details were not published with the announcement.

How is GPT-6 Astra different from earlier ChatGPT models?

The practical difference is what it is built to drive: Dots are persistent agents with their own cloud computer and browser that work in the background, connect to more than 4,000 apps, and can be reached through ChatGPT, mobile, Slack, and Microsoft Teams. Earlier models answered prompts in a chat window, while Astra is presented as the engine for delegated, ongoing tasks.

How can a small business access GPT-6 Astra?

Through ChatGPT Dots, which are rolling out to ChatGPT Pro and Business Premium subscribers in eligible markets, with the first Dot included and extended limits during the first month. Enterprise, education, and healthcare accounts get a beta that an administrator must enable, since it is off by default. Availability varies by region at launch.

Should owners wait for GPT-6 Astra benchmarks before acting?

No, but they should separate the confirmed product facts from the unconfirmed model claims. The agent behaviors, approval controls, and plan availability are documented in launch coverage, while benchmark scores and API pricing were not published. A sensible move is to pilot one delegated workflow now and revisit model-level choices when verified numbers appear.

Next step

If one recurring research or preparation task eats hours every week, that is your pilot candidate. The free six-step AI automation plan on our homepage turns it into a scoped, supervised workflow: start your AI automation plan. If you would rather talk through which task to delegate first, book a call.