OpenAI’s GPT-6 Astra Introduces Human-Level Computer Control

OpenAI has just announced the launch of GPT-6 Astra, its latest flagship artificial intelligence model that brings a massive leap forward in reasoning, alignment, and autonomous computer use.

Available starting today for select enterprise partners, Astra will roll out to ChatGPT Plus, Pro, Business, and Enterprise subscribers over the coming days, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock.

According to OpenAI, Astra brings together years of research in pre-training, reinforcement learning, and safety alignment to create its most capable model to date. Unlike incremental updates, Astra sets new records across standard AI benchmarks. It scored a stunning 99.9% on the ARC-AGI-3 reasoning evaluation and hit 98% on Tier 4 of FrontierMath, a test designed to challenge advanced mathematical reasoning.

The headline feature of GPT-6 Astra is its ability to operate a computer just like a human operator. The model can navigate desktop interfaces, interact with complex software tools, write and run code, and execute multi-step workflows.

During OpenAI’s demonstrations, Astra converted a schematic diagram into a fully printable circuit board (PCB) layout inside KiCad, drafted 3D structural CAD models, created a playable 3D kart racing game, and designed detailed multi-slide presentations. Beyond creative and technical tasks, Astra can autonomously locate pediatricians, file Form 1040 tax documents, browse the web, organize calendars, and execute QA tests on web applications.

OpenAI has also put heavy emphasis on safety and alignment with Astra. When presented with impossible or malicious prompts, earlier models often attempted unauthorized actions or exceeded their scope. In tests mimicking real-world system breaches, GPT-6 Astra respected boundaries in 100% of cases, avoiding unauthorized targets entirely.

Despite its huge performance gains, Astra operates with significantly better token efficiency, cutting API run costs by up to 30% compared to competing models like Claude Fable 5.1 and Claude Opus 5.