OpenAI released GPT-6-Astra in limited preview on 3 September 2026, with a wider rollout the next day. OpenAI president Greg Brockman called it a “generational leap”. It is now part of the line-up inside iOVA, Evosion’s multi-model AI intelligence. Here’s what Astra is, why it’s in the line-up, and what that means for Australian businesses working with iOVA.
What is GPT-6-Astra?
GPT-6-Astra is OpenAI’s newest frontier model. OpenAI’s own one-line pitch is: “Anything you can do on a computer, Astra can do for you. Fast.” It is built for long, multi-step work. That means using a computer and a browser, writing and changing code, and seeing a job through rather than just answering a question.
Key points from the launch coverage (Axios, CNBC):
- Rollout: a staged rollout across ChatGPT, Codex and the OpenAI API, starting with organisations in OpenAI’s cybersecurity programme.
- Scale: OpenAI says Astra was built on its largest-ever training run.
- Benchmarks: OpenAI reports state-of-the-art results on FrontierMath Tier 4, ARC-AGI 3 and TerminalBench-4.0.
- Safety: it is the first OpenAI model rated at the “critical” cybersecurity level under the company’s preparedness framework. The public version refuses advanced cyber tasks such as proof-of-concept exploits.
Why GPT-6-Astra is in iOVA’s line-up
iOVA was never meant to be a single model. He is one intelligence that sends each step of a job to the model best suited to it, then has a different model check the work.
Astra is built for exactly the kind of long, hands-on work iOVA does every day inside real systems such as servers, code repositories, content management systems and ad accounts. It now sits alongside Anthropic’s Claude family, OpenAI’s Codex and Google’s Gemini models (through Google Antigravity) as one of the models iOVA can hand a step to.
How Astra fits into an iOVA job
Every iOVA job runs the same four moves, and Astra plugs into the middle of them:
- Plan. iOVA breaks the task into steps, drawing on what he knows about your business.
- Route. Each step goes to the model best suited to it. A deep planning step might go to Claude Opus 5.5. A hands-on build step can now go to GPT-6-Astra.
- Execute. The work happens directly in your real systems, not in a chat window.
- Verify. A different model reviews the output, and evidence is collected before anything is reported as done. No model approves its own work, including Astra.
A stronger model makes fewer mistakes, but it still makes some. That’s why iOVA cross-checks every output, whichever model produced it.
What it means for your business
- Stronger hands-on builds. DEVY, iOVA’s full-stack developer agent, works with iOVA’s coding tools, and iOVA can route build steps to Astra.
- Better multi-step work. Astra is designed for computer and browser use, which suits jobs that run across several systems.
- No lock-in. You don’t have to pick a winner in the AI race. As new frontier models ship, iOVA’s line-up is updated.
DEVYFrequently asked questions
Has iOVA stopped using Claude or Gemini?
No. Astra adds to the line-up and replaces nothing. Claude, Codex and Gemini each still handle the kinds of work they are best at.
Do I choose when GPT-6-Astra is used?
No. iOVA decides which model handles each step. You set the outcome; he picks the model.
Is my business data used to train GPT-6-Astra?
That depends on your setup, so it isn’t something we can answer in a blog post. Talk to the Evosion team about data handling for your business.
See iOVA’s full AI model line-up for October 2026, or explore iOVA and book a strategy call with Evosion.
iOVA is an Evosion product. GPT, Claude, Gemini and ElevenLabs are trademarks of their respective owners. Evosion is not affiliated with or endorsed by them.
