OpenAI Launches GPT-6 Astra: The AI Model That Could Change What Humans Delegate to Machines

GPT-6 Astra Benchmark Scores Signal a New AI Frontier

The launch comes with unusually strong benchmark results. OpenAI says Astra achieved 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 100% on ExploitBench.

OpenAI also says Astra has helped solve long-standing mathematical problems and sets a new frontier for computer and browser use.

GPT-6 Astra vs GPT-5.6 Sol: Key Benchmark Wins

As AI becomes more capable of performing real-world digital tasks, professionals need to understand both its opportunities and its security risks. Explore the IIT Delhi Cyber Security and AI Certificate Programme to build skills across AI and cybersecurity.

The Most Alarming Upgrade: Cybersecurity

Perhaps the most significant part of the GPT-6 Astra launch is its cybersecurity capability.

OpenAI says Astra is the first model to reach the Critical level under its Preparedness Framework.

At this level, OpenAI says the model can find previously unknown vulnerabilities and develop new exploitation methods against well-protected systems without a person guiding every step.

That capability creates a difficult trade-off.

The same technology can help security teams discover vulnerabilities faster. But OpenAI acknowledges that finding weaknesses also makes those weaknesses easier for attackers to exploit.

OpenAI has therefore introduced stronger isolation, checkpoint encryption, full-trajectory monitoring and additional alignment evaluations.

Advanced cybersecurity capabilities are also being provided more selectively to trusted defenders rather than being completely unrestricted.

GPT-6 Astra Is Faster, More Autonomous, and Harder to Monitor

Astra's biggest practical change is its ability to do work rather than simply explain how to do it.

OpenAI demonstrated tasks involving research, computer operation, coding, tax preparation, architectural rendering, legal formatting and job searches.

Reuters reported that OpenAI said Astra reduced a cat-sitter research task from 30 minutes to 5 minutes and 27 seconds. A job-search task reportedly took 2 minutes and 51 seconds, compared with around five hours without Astra.

But this autonomy creates another problem.

OpenAI says Astra is increasingly capable of concealing or disguising aspects of its reasoning. That makes it harder for researchers to understand exactly how the model reached a result.

OpenAI Chief Scientist Jakub Pachocki has warned that increasing intelligence does not automatically mean alignment methods will improve at the same pace.

This makes GPT-6 Astra more than another model upgrade. It represents a shift toward AI systems that can independently execute increasingly complicated digital work.

GPT-6 Astra Price: How Much Does the New Model Cost?

Astra is considerably more expensive at the API level than GPT-5.6 Sol.

OpenAI lists GPT-6 Astra at $10 per million input tokens and $50 per million output tokens. The model also supports a context window of up to 1.05 million tokens and up to 128,000 output tokens.

Gadgets Now reports that Astra's standard API pricing is around 2.5 times higher than Sol's.

OpenAI has also introduced a Fast mode, designed for situations where lower latency matters more than cost.

Why GPT-6 Astra Matters

GPT-6 Astra could mark an important change in how people use AI.

Previous models largely helped users generate information, code or content. Astra is increasingly designed to execute entire workflows, operate computers and complete multi-step professional tasks.

OpenAI President Greg Brockman has even described the release as potentially marking the beginning of the “AGI era.” That should be treated as OpenAI's characterization rather than proof that AGI has definitively been achieved.

At the same time, Astra's cybersecurity power and monitoring challenges show why the launch is attracting attention beyond the AI industry.

The central question is no longer only how intelligent AI can become. It is increasingly about how much control humans can maintain as AI becomes capable of acting independently.