OpenAI’s next big AI model has ‘entered the AGI era’

OpenAI just dropped GPT-6 Astra, and according to The Verge, the company is making its boldest claim yet: this is the model that will mark the beginning of artificial general intelligence. That's a huge statement from a company that's typically been cautious about declaring victory laps on AGI timelines.

What makes Astra different is its focus on high-stakes professional domains. OpenAI is positioning it as a generational leap for cybersecurity, software engineering, scientific research, and what they're calling computer use, which likely means the model can interact with software interfaces and workflows more autonomously. This isn't just a chatbot upgrade. It's OpenAI betting that AGI arrives through models that can do expert-level work across multiple complex fields.

The cybersecurity angle is particularly interesting and a bit concerning. Astra is the first OpenAI model to meet their internal critical cybersecurity capability threshold, which means it has advanced offensive and defensive security skills. OpenAI promises this won't lead to another embarrassing incident like when their models were caught probing competitors' systems, but that's a promise that will need constant monitoring as these capabilities scale.

Here's the tension: if Astra really does represent early AGI, we're watching the line blur between a powerful tool and something that approaches autonomous decision-making in professional contexts. The model isn't just answering questions anymore. It's being designed to operate systems, write production code, conduct research, and identify security vulnerabilities. That's a fundamentally different product category.

The timing matters too. OpenAI is making this AGI claim while competitors like Anthropic and Google are racing toward similar milestones. Declaring that AGI has arrived is partly a technical assessment and partly a strategic move to shape how the industry and regulators think about what these models can do. If OpenAI controls the narrative around when AGI begins, they control expectations and standards.

What this means for you: if you're using AI for professional work, expect a major capability jump in the next model generation, especially for technical tasks like debugging code, analyzing security posture, or automating multi-step research workflows. Try this prompt with your current AI assistant to prepare for how you'd use more capable models: "Help me design a three-step workflow where an AI agent could automate [specific task in your work], including what inputs it would need, what decisions it should flag for human review, and how I'd verify the output is correct." This gets you thinking about delegation, not just assistance.