5 Key Takeaways
- OpenAI released GPT-6 Astra, its most intelligent and aligned model, designed for real-world computer work, software engineering, science, cybersecurity, and professional tasks.
- Astra is rolling out first to limited organizations, then to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock; GPT-6 Astra Pro is available for selected paid plans.
- Astra focuses on computer use, including filling forms, updating customer records, organizing calendars, conducting research, analyzing data, building websites, installing and testing software, and troubleshooting; it scored 72.6% on OSWorld 2.0.
- Astra improves software engineering and professional work, scoring 57.9% on Terminal-Bench 4.0 and creating documents, presentations, and spreadsheets that follow user formatting and style.
- Astra shows major science, math, and cybersecurity gains, including 96% on GPQA Diamond, 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench without production safeguards, while refusing some advanced cybersecurity requests and using stronger safeguards.
OpenAI Releases GPT-6 Astra: A New AI Model Built for Real-World Computer Work
OpenAI has officially introduced GPT-6 Astra, its most intelligent and aligned artificial intelligence model to date. The new model is designed to handle computer use, software engineering, science, cybersecurity, and professional work. OpenAI says Astra is rolling out first to a limited group of organizations. In the coming days, it will become available to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock. The company has also introduced GPT-6 Astra Pro for selected paid plans.
This latest release signals a shift beyond chatbots toward AI systems that can operate software and complete tasks directly on a computer. Instead of only answering questions or generating text, GPT-6 Astra is meant to interact with screens, work with files, and execute multi-step workflows. OpenAI says the model is its most capable yet in professional and technical settings. The initial limited rollout is intended to let organizations test the system before broader availability.
A Model Built for Computer Use
One of Astra’s central focuses is computer use. OpenAI says GPT-6 Astra can carry out tasks on computers including filling online forms, updating customer records, organizing calendars, and conducting online research. It can also work with software to analyze scientific data, create plots, build websites, and run frontend tests. Beyond that, OpenAI says the model can install and test software and help troubleshoot problems shown on a computer screen. These abilities point toward a model that can act more like a digital assistant for day-to-day technical and administrative work.
On the OSWorld 2.0 benchmark, which measures how well AI models complete real computer tasks, Astra scored 72.6%. It completed tasks in about 40 minutes. By comparison, OpenAI’s earlier GPT-5.6 Sol scored 65.7% and needed about 75 minutes. OpenAI adds that Astra completed tasks on the Mind2Web benchmark 1.9 times faster when combined with an updated Codex system. Benchmarks like these are standardized tests used to compare AI systems on specific tasks, and they give a rough sense of how models perform outside controlled lab conditions.
Stronger Software Engineering and Professional Work
OpenAI says GPT-6 Astra is its best model for software engineering to date. On Terminal-Bench 4.0, which tests complex terminal-based tasks, Astra scored 57.9%. That compares with 37.3% for GPT-5.6 Sol and 55.8% for Claude Fable 5.1 in the comparison shared by OpenAI. The benchmark focuses on difficult command-line work, debugging, and system-level software tasks, making the higher score notable for developers and technical teams.
Beyond coding, GPT-6 Astra is designed to create documents, presentations, spreadsheets, and other work based on existing templates. OpenAI says the model can follow a user’s formatting and style while avoiding unnecessary information in the final output. That means it is aimed at professionals who need polished materials that match existing company or personal formats. The model is intended to reduce the manual effort involved in preparing routine business documents and reports.
Science, Math, and Cybersecurity Gains
Astra also improves performance in scientific and mathematical tasks. OpenAI reports a 96% score on GPQA Diamond, a test covering graduate-level biology, chemistry, and physics questions. Astra scored 98% on FrontierMath Tier 4 and 99.9% on ARC-AGI-3. These are demanding evaluations that test advanced reasoning and problem-solving. OpenAI says Astra has already helped solve long-standing open problems in mathematics.
Cybersecurity remains a major focus for the new model. OpenAI says GPT-6 Astra represents a major increase in cybersecurity capabilities. The model scored 100% on ExploitBench without production safeguards, compared with 78.5% for GPT-5.6 Sol. On ExploitGym, Astra achieved a 42.4% success rate compared with 30.3% for GPT-5.6 Sol. These tests measure a model’s ability to find and exploit software vulnerabilities in controlled settings.
OpenAI says Astra can identify vulnerabilities and help defenders with tasks such as secure code review and patching. However, the version launching now will refuse some advanced cybersecurity requests, including creating proof-of-concept exploits for vulnerabilities. The company also says it has added stronger safeguards, monitoring systems, and protections against attempts to bypass security controls. This dual approach is designed to support defensive security work while limiting potential misuse.
Price and Availability
GPT-6 Astra is initially being rolled out to a limited number of organizations. OpenAI says it will become available to ChatGPT Plus, Pro, Business, and Enterprise users over the coming days. Astra will also be available through the OpenAI API, Microsoft Azure, and AWS Bedrock. This means both direct ChatGPT subscribers and enterprise cloud customers will be able to access the model.
For developers using the OpenAI API, standard pricing is $10 per million input tokens and $50 per million output tokens. In the AI context, tokens are pieces of text or code that models process and generate. OpenAI says Fast mode will offer up to twice the speed of standard processing at twice the standard price. Enterprise administrators will be able to enable Astra for their workspaces, with access turned off by default at launch. OpenAI also says eligible API customers can use Zero Data Retention.
What Comes Next
The staged launch suggests OpenAI is prioritizing safety and controlled access before wider distribution. Organizations using Microsoft Azure and AWS Bedrock will be able to integrate Astra into their existing cloud workflows. As the model reaches more ChatGPT users, the full impact on professional productivity, software development, and cybersecurity work will become clearer. For now, GPT-6 Astra stands as OpenAI’s most ambitious attempt to move AI from a conversational tool into an active worker on computers.
No comments:
Post a Comment