OpenAI’s Astra helps turn ideas into working software

Having an idea is one thing. Knowing your way around the software needed to build it is another.

OpenAI’s GPT-6 Astra, announced on 3 September, is designed to bridge more of that gap. It can work across applications, write and test code, and produce documents or interactive projects from instructions.

In OpenAI’s reported testing, Astra scored 72.6% on an evaluation of computer tasks, compared with 65.7% for its previous frontier model. These are the developer’s evaluation results, rather than a guarantee that every real-world task will succeed.

Early user Matt Shumer offered a more playful demonstration: a world built in Unreal Engine, populated by AI agents that coordinated to survive. NotBoring highlighted it as an example of the creative possibilities.

The potential benefit is making complex tools more approachable and reducing the repetitive work between an idea and a usable result. Stronger capabilities also bring risks: OpenAI has restricted advanced cybersecurity uses and added monitoring for unauthorised actions. This is promising progress, with human judgement still needed to check the work.