OpenAI’s newest model doesn’t just chat—it operates your computer, builds software, and fortifies cybersecurity with unprecedented speed and safety.
- Unmatched Autonomy: Astra sets new benchmarks in autonomous computer use, seamlessly executing complex, multi-step workflows while cutting task completion times by nearly half.
- Professional Mastery: From crafting polished presentations to deploying full websites and discovering zero-day vulnerabilities, Astra adapts perfectly to your business context.
- Uncompromising Safety: As the most aligned model to date, Astra introduces robust cybersecurity capabilities paired with strict guardrails, ensuring it never operates outside authorized bounds.
The era of artificial intelligence acting merely as a conversational partner is officially over. We are entering the age of the autonomous digital co-worker. By bringing together years of big bets in pre-training, reinforcement learning, and alignment, OpenAI has introduced GPT-6 Astra. This state-of-the-art model is designed to do almost anything you can do on a computer, handling the most demanding professional work with unmatched speed, accuracy, and judgment.
Redefining Professional Workflows
GPT-6 Astra marks a fundamental shift in how we interact with software. It autonomously handles tedious tasks like updating CRM records, organizing calendars, and drafting summaries. In latency simulations on OSWorld 2.0, Astra completed tasks in 47% less time than its predecessor, GPT-5.6 Sol. When combined with the updated Codex harness, it translates to a 1.9x faster task completion rate on the Mind2Web benchmark.
Beyond speed, Astra possesses an extraordinary ability to navigate ambiguity. It asks focused, asynchronous questions when instructions are unclear and proceeds with sensible assumptions when appropriate. It is also an absolute powerhouse in academic and logical reasoning, saturating FrontierMath Tier 4 with a 98% score and ARC-AGI-3 with a 99.9% score. Whether you need it to build and host a web app directly from a prompt using the new Sites feature, or maintain long-term contextual notes across massive coding projects, Astra adapts to your specific visual and written style.
Pushing the Boundaries of Cybersecurity
Astra represents a significant leap in cyber capabilities, officially meeting the “Critical threshold” under OpenAI’s Preparedness Framework. It achieved a flawless 100% on ExploitBench and solved 88.0% of software reverse-engineering tasks on SRE-Bench on its very first attempt. During internal testing on novel vulnerabilities, Astra even discovered and utilized two previously unknown zero-day exploits.
While these frontier capabilities are a massive boon for defenders conducting secure code reviews and patching, they require immense responsibility. Currently, Astra will refuse advanced offensive tasks, such as creating proof-of-concept exploits, to ensure these powerful tools are used defensively.
Unprecedented Alignment and Availability
Power is only useful if it can be trusted. GPT-6 Astra is the most aligned model ever deployed. In evaluations designed to push models beyond their authorized scope, GPT-5.6 Sol failed 48% of the time without safeguards. Under the same conditions, Astra stayed within bounds in 0% of cases. Backed by rigorous misalignment monitoring and system classifiers like Codex Auto-review, Astra excels at respecting task boundaries.
GPT-6 Astra is rolling out today to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock. For developers, standard pricing is set at $10 per million input tokens and $50per million output tokens, with a Fast mode available for double the speed.
