OpenAI Releases GPT-6 Astra, Claiming AGI Milestone

OpenAI announced the release of its newest model, GPT-6 Astra, on Thursday, with company president Greg Brockman stating, “Welcome to the AGI era.” OpenAI defines AGI as "highly autonomous systems that outperform humans at most economically valuable work." The company asserts that Astra is its most intelligent and safest model to date, incorporating advancements in pre-training, reinforcement learning, and safety guardrails. According to OpenAI VP of research Aidan Clark, the training process for Astra utilized over 100,000 GPUs. OpenAI reported that Astra surpassed all other frontier models in benchmark tests covering computer use, browser navigation, software engineering, cybersecurity, scientific tasks, and professional work. The model achieved perfect or near-perfect scores on several challenging evaluations, including 98.6% on ARC-AGI 3, a reasoning test designed to resist pre-computation by the model. Brockman further commented, "When we look back and ask when AGI arrived, we’ll see that it’s about this time and it’s about this model." Astra is initially rolling out to enterprise customers with Daybreak access, OpenAI’s restricted program for cybersecurity professionals. It will subsequently become available to Plus, Pro, Business, and Enterprise users, via the OpenAI API, and on Amazon Web Services in the coming days. In terms of computer use, OpenAI stated that Astra can perform tasks such as filling out forms, updating CRM records, conducting research, drafting summaries, analyzing data, building websites, executing frontend QA checks, and troubleshooting on-screen problems. The company also reported that Astra completed more desktop tasks correctly than its predecessor, GPT-5.6 Sol, while operating in approximately half the time. A new software and orchestration layer, known as a harness, enables the Codex coding agent to complete web-based tasks 1.9 times faster. Demonstrations showcased Astra's capabilities, including laying out a circuit board, constructing a business dashboard, and completing a federal tax return in a web browser using a W-2 form. Regarding professional work, the company indicated that Astra represents a significant leap forward. The model's development involved extensive pre-training, reinforcement learning techniques, and the implementation of robust safety guardrails, positioning it as OpenAI's most capable and secure offering to date. The benchmark results, particularly its performance on complex reasoning and professional tasks, are central to OpenAI's claim of achieving AGI with this release. The phased rollout strategy aims to manage access and gather feedback from different user segments, starting with specialized cybersecurity professionals before broader availability.