OpenAI released its new AI model, GPT‑6 Astra, on Thursday evening (Hungarian time). The launch comes roughly one year after GPT‑5 and two months after the most recent incremental release, GPT‑5.6.
What OpenAI claims
According to OpenAI, GPT‑6 Astra is “the world’s best computer‑use model,” able to navigate computer systems autonomously much better than previous AI. In a demonstration video, Astra followed voice instructions to perform multiple simultaneous tasks: it turned a simple shape into a spaceship design, edited legal documents, ordered food and reserved a tennis court. The demonstration also showed the model rapidly switching between programs and understanding fast, mumbled speech.
OpenAI executives signaled the release’s significance. Greg Brockman, president of OpenAI, said it is “not unreasonable to feel that we are now in the AGI era,” adding that in a few years people might look back and declare this the moment AGI arrived. AGI—artificial general intelligence—refers to a system capable of performing any intellectual or cognitive task that a human can.
Safety and transparency: OpenAI’s position
OpenAI emphasized that Astra is the safest model they have built to date, stating that the model’s chain‑of‑thought is kept under tight supervision to prevent misinterpretation and misuse of tasks. The company has an interest in addressing recent concerns around advanced models.
External concerns: opacity and detection of risky behavior
Researchers and technology press have raised cautionary notes. Many current AI models expose internal reasoning to some extent—so‑called “chain‑of‑thought” explanations—which allows researchers and automated safety systems to track model activity and detect problematic behaviors before they cause harm.
Sources familiar with Astra’s development say the new model uses a different, less transparent technique that “mixes” information across internal layers before output. That approach can make it harder to observe and identify potentially dangerous behaviors, and some researchers have voiced fears about catastrophic risks associated with the model’s opacity.
Availability
OpenAI said GPT‑6 Astra would be made available in the days following the announcement to subscribers of ChatGPT Plus, Pro, Business and Enterprise tiers. The company also plans to offer the model via the OpenAI API and through Amazon Web Services.
Why this matters
The debut of GPT‑6 Astra illustrates both rapid advances in large model capabilities—handling complex, multimodal, system‑level tasks—and renewed concerns about how to ensure safety and transparency as models become more sophisticated. OpenAI’s optimistic framing and independent researchers’ warnings are likely to continue shaping debate as the model is rolled out more broadly.
