OpenAI has begun rolling out GPT-6 Astra, the company’s newest and most capable model, on 3 September 2026. It is the first major upgrade to ChatGPT and Codex in over a year.
OpenAI calls Astra its most capable model for complex work. That covers coding, research, computer use and long multistep tasks. Astra can also create documents, spreadsheets and presentations that follow your templates, and adjust when you change instructions midway.
Quick Take
- GPT-6 Astra started rolling out on 3 September 2026.
- A limited set of organisations gets access first, then ChatGPT Plus, Pro, Business and Enterprise users over the coming days.
- OpenAI says it is the first model to reach its “Critical” cyber threshold, so top capabilities stay restricted.
Who gets access first
Access is staged, not open to all on day one. A limited group of firms in OpenAI’s Daybreak Access program gets Astra first. Broader access follows over the coming days for ChatGPT Plus, Pro, Business and Enterprise users, plus the OpenAI API and Amazon Web Services.
For enterprise workspaces, access is off by default at launch. An administrator must switch it on. Pro, Business and Enterprise plans also get a stronger variant called GPT-6 Astra Pro.
What it can do
OpenAI reports large benchmark gains. Astra scored 99.9% on ARC-AGI-3 and 100% on ExploitBench, per the company. It also improves computer use, letting ChatGPT operate a Mac or other desktop, including in the background while you work.
Codex gets a specific upgrade too. Astra can keep notes across context windows and search earlier messages, instead of compressing long sessions into one summary that drops detail. The feature is experimental for now and sits behind a config setting.
The safety catch
Here is the part builders should read twice. OpenAI says Astra is the first model to cross the “Critical” cybersecurity level in its Preparedness Framework. In company tests, it developed exploits for hardened browsers and operating systems, and found two previously unknown vulnerabilities.
OpenAI had already slowed Astra’s release to add safety testing. The company also said Astra was harder to monitor than its earlier Sol model in evasion tests. President Greg Brockman went further, telling reporters, “Welcome to the AGI era.”
What this means for you: If you build on OpenAI, plan for higher per-token cost and staged access, and check which Astra capabilities your plan actually unlocks before you promise features to your own users.
StartupFeed Insight
Astra is priced as premium intelligence, not a commodity. Early reports put API cost near $10 per million input tokens and $50 per million output tokens, roughly double OpenAI’s own GPT-5.6 Sol at $5 and $30. OpenAI’s bet is that stronger reasoning means fewer tokens and fewer failed attempts, so the finished job costs less even at a higher sticker price. For Indian startups running tight margins, that maths only works if you measure cost per completed task, not cost per token. Watch for the API and AWS availability date, since that is when Indian teams can actually test it. Expect broad ChatGPT access within days of this launch.
— Avinash Mishra, Business Correspondent
