OpenAI ChatGPT-6 Astra Release: Enhanced Computer Skills and Technical Thinking

Autonomous Computer Use and Workflow Automation

GPT-6 Astra is ushering in a significant transformation in the way AI interacts with digital environments. The model is capable of performing complex tasks in various software applications, including updating CRM records, managing calendars, and conducting online research. In technical fields, the model can edit printed circuit board (PCB) layouts in KiCad, generate website front ends, and perform quality assurance checks.

Performance data indicates a significant increase in efficiency. In simulations on OSWorld 2.0, the model completed tasks in approximately 40 minutes, compared to 75 minutes for GPT-5.6 Sol. Furthermore, task completion speed in the Mind2Web benchmark has improved by a factor of 1.9 compared to earlier iterations.

Professional and Scientific Skills

The model demonstrates a high level of competence in specialized professional workflows. In 3D design, it can create models in Blender and convert them into walkable scenes within Unreal Engine 5. For corporate environments, it is trained to adhere to specific business templates when creating presentations and spreadsheets.

In the academic and scientific fields, GPT-6 Astra has achieved high benchmark scores:

• Mathematics: 98% on FrontierMath Tier 4.
• Abstract Thinking: 99.9% on ARC-AGI-3.
• Scientific Reasoning: 96% on the GPQA Diamond benchmark, which tests graduate-level knowledge in biology, chemistry, and physics.

Software Development and Programming

For software development, GPT-6 Astra offers improvements in both output quality and context management. On Terminal Bench 4.0, the model achieved an accuracy rate of 57.9%, which is significantly higher than the 37.3% recorded by GPT-5.6 Sol.

A new feature in the Codex Harness enables the model to preserve and retrieve context across different windows. Instead of relying on traditional compression and summarization methods, which can result in the loss of details, Astra manages searchable notes across context windows, ensuring that previous requirements and test results remain accessible during long debugging or refactoring sessions.

Cybersecurity and Security Focus

The model demonstrates advanced capabilities in cybersecurity and achieves a 100% score on the ExploitBench. In internal tests without production safeguards, the model identified two previously unknown zero-day vulnerabilities. It also demonstrated a high proficiency in reverse engineering software binaries and solved 88% of the SRE Bench tasks on the first attempt.

In terms of safety and alignment, the model exhibits a reduction in “capability hallucinations,” meaning it is less likely to make inaccurate statements about its own capabilities. In a stress test on computer usage safety, the misalignment rate was 2.4%, compared to 22% for GPT-5.6 Sol.

Additionally, when evaluating impossible tasks, the model showed a 0% rate of attempts to circumvent authorized target limits.

ChatGPT 6 Availability and API Pricing

GPT-6 Astra will be integrated into existing subscription tiers. Enterprise administrators can enable the model for their workspaces, although it is disabled by default at launch.

For developers using the OpenAI API, the model is available under the identifier gpt-6-astra. Standard pricing is $10 per million input tokens and $50 per million output tokens. A “Fast” mode is also available, offering double the processing speed at double the standard price.