GPT-6 Astra: OpenAI’s Most Advanced Model Yet — Here’s What It Can Actually Do

OpenAI has just pulled back the curtain on GPT‑6 Astra, and if the benchmark numbers are anything to go by, this isn’t an incremental update — it’s a genuine leap. Positioned as the company’s most intelligent and aligned model to date, Astra pushes the boundaries of what AI can autonomously handle, from writing production-ready code to conducting scientific research and even navigating a computer like a human would.

Here’s a breakdown of what’s new, what it means, and why it matters.

What Makes GPT-6 Astra Different

OpenAI describes Astra as state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work. That’s a broad claim, but the benchmark results back it up in several key areas.

A Genuine Leap in Computer Use

One of Astra’s headline features is how it interacts with actual software environments rather than just generating text. It can take care of tedious tasks like filling out online forms, updating customer records in a CRM, and organizing calendars, conduct online research, analyze scientific data, generate plots, create websites, and run frontend QA checks.

The efficiency gains are significant too. In latency simulations on OSWorld 2.0, Astra achieves higher computer-use performance in about 47% less time per task than its predecessor, scoring 72.6% at roughly 40 minutes per task, compared with 65.7% at roughly 75 minutes.

Built for Professional Workflows

Beyond raw benchmarks, Astra is designed to slot into real business environments. It creates clear, well-structured documents, presentations, spreadsheets, and analyses that follow existing templates and match a user’s writing and visual style, while specifically pulling only the context that matters into outputs instead of repeating unnecessary information.

Industry partners have already weighed in. Niko Grupen, Head of Applied Research at Harvey, noted that Astra approaches legal work the way a discerning lawyer does — distinguishing documents from established records, surfacing unsupported assumptions, and converting gaps into concrete drafting positions.

The Coding Upgrade

For developers, Astra represents what OpenAI calls its best model for software engineering yet. John Crepezzi from Jane Street’s AI Assistants team said Astra delivers state-of-the-art performance on their internal coding benchmarks, communicates in a way that’s easier for developers to follow, and produces code that requires less iteration to reach production quality.

Astra also introduces smarter memory handling during long coding sessions. Rather than repeatedly compressing context into a single summary, it can keep notes across context windows while preserving accumulated details, and earlier context windows remain searchable.

Cybersecurity: Powerful, But Tightly Controlled

This is where things get serious. Astra is a significant jump in cyber capabilities and meets the Critical threshold in cybersecurity under OpenAI’s Preparedness Framework — its ability to identify and develop zero-day exploits can help defenders find and patch weaknesses, but also creates a need for stronger safeguards.

Notably, during evaluation, Astra discovered and used two previously unknown zero-day vulnerabilities, which OpenAI is disclosing to their respective maintainers. As a result, OpenAI has built in restrictions: Astra will refuse to comply with more advanced cybersecurity tasks such as creating proof-of-concept exploits for vulnerabilities, with expanded access planned for vetted defensive use cases.

Alignment and Safety Improvements

OpenAI is positioning Astra as its most aligned model so far. Compared to its predecessor, which went beyond an authorized target 48% of the time in a difficult-task evaluation, GPT‑6 Astra did this in 0% of cases.

The model also appears more transparent about its own limitations. In one evaluation, Astra is three times less likely than its predecessor to make inaccurate representations about its capabilities and affordances.

GPT-6 Astra: Availability and Pricing

GPT‑6 Astra is available in the OpenAI API as gpt-6-astra, priced at $10 per million input tokens and $50 per million output tokens, with a Fast mode offering up to 2x the speed at 2x the standard price. It also supports Zero Data Retention for eligible API customers.

Quick Recap

Curious how rapid advances in AI models are reshaping content creation, sports analytics, and gaming journalism? Stay ahead of the tech curve with more breakdowns on TechnoSports.