OpenAI has released GPT-6 Astra and declared the arrival of the "AGI era," marking a turning point the company has avoided claiming for years. President Greg Brockman's statement signals OpenAI's belief that this model represents a genuine shift toward artificial general intelligence rather than incremental improvement over previous systems.
Astra sets new performance records across multiple demanding domains. The model dominates benchmarks in mathematics, coding, and cybersecurity. It stands apart from prior OpenAI releases in one critical measure: it is the first model OpenAI classifies as "critical" under its own safety framework. This designation reflects the company's internal assessment that Astra poses risks requiring elevated precautions.
The most striking evidence of Astra's capabilities emerged during safety testing. The model independently discovered two previously unknown zero-day vulnerabilities without explicit instruction to do so. Zero-day exploits represent the most dangerous class of security flaws because vendors have had zero days to patch them. Finding novel vulnerabilities requires not just pattern matching but genuine reasoning about system architecture and attack vectors. This discovery happened without the researchers directing Astra toward vulnerability hunting, suggesting the model developed instrumental understanding of cybersecurity principles.
OpenAI's willingness to explicitly claim AGI arrival differs sharply from prior statements. The company previously dodged such declarations, framing models like GPT-4 as "advanced" but stopping short of the AGI label. The shift reflects either confidence in Astra's capabilities or a recalibration of what OpenAI considers the AGI threshold. The company's formal safety framework now includes a "critical" tier above previous classifications, implying the system poses novel risks requiring new protective measures.
The Astra release raises immediate questions about deployment and access. OpenAI has not disclosed whether Astra will be made available through ChatGPT's free tier, as with prior models. Given the "critical" safety classification, the company may restrict access to enterprise customers or require additional authentication and usage monitoring. The discovery of zero-days during testing suggests OpenAI conducted extensive red-teaming before release, though details on those efforts remain sparse.
The AGI claim carries business and regulatory weight. If OpenAI publicly designates Astra as AGI-class, that framing will shape how governments, competitors, and investors interpret the technology's trajectory. It influences narratives around AI regulation, raises questions about OpenAI's governance structure and whether it meets its founding charter commitments around safety, and affects competitive dynamics with Anthropic, Google DeepMind, and other labs.
The mathematics and coding benchmarks reflect concrete capabilities relevant to professional work. Excellence in these domains positions Astra as a tool for engineers, researchers, and developers. The cybersecurity performance carries different implications. A model that finds zero-days on its own could accelerate both defensive and offensive capabilities in security research. That capability alone justifies OpenAI's "critical" safety tier.
Astra represents the clearest signal yet that AI systems are approaching capabilities traditionally assumed to require human-level general reasoning. Whether this constitutes true AGI depends on how one defines the term. OpenAI's declaration that we have entered the "AGI era" reflects confidence that the line between narrow task performance and general-purpose reasoning has been crossed.