Model

GPT-6 Astra: The Model Behind OpenAI Dots

GPT-6 Astra is the frontier model under every Dot. Here is what OpenAI has said about it — and, just as important, what it has not.

Last verified: September 30, 2026 · Reviewed by AstraDot editorial desk

GPT-6 Astra is the frontier model that powers OpenAI Dots. When you give a Dot a project, Astra is the intelligence doing the reasoning, the planning, and the tool use behind the scenes. It is the second half of the Dots story: the Dots platform supplies the body — a cloud computer, a browser, connection to apps — and Astra supplies the thinking.

What GPT-6 Astra is

OpenAI names GPT-6 Astra as the model behind Dots, and describes Dots as agents with their own cloud computer that can write and test code, browse, connect to more than 4,000 apps, and learn from feedback over time. Astra is the model that makes those behaviors possible. It is positioned as a frontier model — the class OpenAI builds for its most demanding reasoning and agentic work.

That is the full extent of the durable, on-the-record claim. OpenAI's launch material is a product announcement, not a model card, so it tells you what Astra does in service of Dots rather than how it scores against other models.

Why a frontier model matters for always-on agents

A one-shot chat reply is forgiving: if the answer is weak, you rephrase and try again. An always-on agent is not. It may run for hours, touch real systems, and make dozens of small decisions on your behalf. That raises the bar on three things in particular:

  • Reasoning. The agent has to decompose an ambiguous goal — "turn this feedback into fixes" — into concrete steps it can execute and verify.
  • Action. It has to use tools correctly: drive a browser, run code, call connected apps, and recover when something fails.
  • Long-horizon work. It has to stay coherent across a project that spans many steps and interruptions, including while handling several projects.

A model that is weak at any of these produces an agent that drifts, loops, or breaks things. That is why OpenAI frames Dots as a frontier-model product: the agent's usefulness is capped by the model's ability to reason and act reliably over time.

What OpenAI has said — and what is not published

This is the section to read carefully, because it is where most coverage overreaches.

ClaimStatus
GPT-6 Astra powers OpenAI Dots Published — stated by OpenAI in the Dots announcement
Dots have their own cloud computer, browser, and code execution Published — part of the Dots description
Benchmark numbers for GPT-6 Astra Not published
A detailed model card or eval table for Astra Not published — OpenAI links a change log instead
Context window size, training details, or pricing per token Not published

In short: there are no benchmark numbers published for GPT-6 Astra in the Dots launch material. OpenAI instead maintains an official GPT-6 Astra change log, which is the right place to check for its own updates. This page deliberately does not quote scores from third parties, because we cannot verify them.

How Dots use GPT-6 Astra

The model never acts in a vacuum. A Dot wraps Astra in a specific environment and a specific workflow:

  • Its own cloud computer. Astra operates inside an isolated machine, separate from your devices unless you connect them. That gives the model a place to run code, keep files, and recover from mistakes without touching your own setup.
  • A browser and apps. Astra can browse and act through connected apps, which is how a Dot moves from "writing about work" to "doing work".
  • Feedback over time. A Dot learns from your corrections, so Astra's behavior on your project is meant to improve as you steer it.

Safety and oversight context

A capable model doing real work needs guardrails, and OpenAI describes several that sit around Astra rather than inside it. Background proactive research uses read-only tools, so a Dot cannot send messages, change app content, or control your browser or computer while researching. An auto-review step checks actions against your instructions, Custom Rules, and safety requirements. Some sensitive tasks, such as changing a password, stay with you. OpenAI can also pause or stop a Dot if monitoring finds a concern. Our security and privacy page covers these in detail.

A frontier model is still a model that can make mistakes. Review consequential work before you act on it.

How to read claims about Astra skeptically

Because Astra is the model behind a hyped product, it attracts confident-sounding claims. A few habits will keep you grounded:

  1. Ask for the source. If a number is not in OpenAI's own documentation, treat it as unverified. The official change log is the anchor.
  2. Separate the model from the product. A Dots feature — say, a Slack integration — is not an Astra capability on its own. Both are real, but they are different claims.
  3. Prefer behavior over leaderboards. For agentic work, what matters is whether a Dot finishes your task reliably, not a single benchmark delta.
  4. Watch for invented dates and prices. OpenAI has not published a date for wider Dots availability or a price for additional Dots; anything specific is a guess. See Dots pricing.

Frequently asked questions

What is GPT-6 Astra?
GPT-6 Astra is the frontier AI model that powers OpenAI Dots. OpenAI identifies it as the model behind Dots, but has not published a separate benchmark report or capability scorecard for it as part of the Dots launch.
Has OpenAI published GPT-6 Astra benchmark numbers?
No. Benchmark numbers for GPT-6 Astra are not published in the Dots launch materials. OpenAI points to its model documentation and change log; we link the official GPT-6 Astra change log rather than repeat unofficial figures.
What makes GPT-6 Astra suited to always-on agents?
An always-on agent has to reason about a goal, act on it, and stay coherent over a long horizon. A frontier reasoning model is what lets a Dot plan multi-step work, use tools such as a browser and a code editor, and keep a project moving over time. OpenAI has not published the specific scores behind those claims.
Does every Dot use GPT-6 Astra?
Yes. OpenAI states that Dots are powered by GPT-6 Astra. A Dot uses that model inside its own cloud computer and learns from the feedback you give it over time.
Should I trust benchmark claims about Astra from other sites?
Be skeptical. If a number is not in OpenAI's own documentation, treat it as unverified. This page only repeats what OpenAI has published and marks everything else as not published.