OpenAI has unveiled its latest flagship model, GPT-6 Astra, which the company considers a significant milestone in the journey towards Artificial General Intelligence (AGI). According to Greg Brockman, co-founder and president of OpenAI, it is no exaggeration to state that we are entering the AGI era, and future historians may refer to this moment as the beginning of that transformation.
Astra stands out for its impressive capabilities, including the ability to operate independently with computers and browsers, write code, and tackle complex mathematical problems. In tests, the model outperformed the average human by quickly scheduling appointments with the U.S. Department of Motor Vehicles, searching for job openings, and finding housing options.
Access to Astra will roll out in the coming days, initially for corporate clients enrolled in the early access program, Daybreak, followed by subscribers of ChatGPT Plus, Pro, and Business. Currently, there are no plans for free access. For the premium tiers, strict message limits have been set: ChatGPT Pro at $200 allows 200 messages per week with Astra Pro, while the $100 tier permits only 50 messages. Notably, the 50-message quota applies collectively to both GPT-6 Pro and GPT-5.6 Sol Pro.
In a recent test dubbed BenchCAD, which assesses how well models reconstruct 3D objects from images taken from various angles by generating code, GPT-6 Astra achieved the highest score, with a geometric match of 95.9%. This surpasses GPT-5.6 Sol’s 83.3% and Claude Fable 5.1.5’s 84.3%. The estimated cost of queries in these configurations is approximately 43% lower than that of Sol and 86% lower than Fable 5.1.
Furthermore, Astra's capabilities in cybersecurity have notably expanded. OpenAI has classified the model's performance as critical, stating that with the right tools and access, Astra can autonomously identify previously unknown vulnerabilities and devise exploitation methods for secured systems.
In anticipation of the release, OpenAI has enhanced Astra's security measures by implementing stricter limitations on hazardous cyber requests, employing system classifiers, and establishing multi-layered monitoring protocols.
However, OpenAI acknowledges the growing challenge of controlling such powerful models. Chief Scientist Jakub Pachocki remarked that monitoring Astra's internal reasoning could hinder further scaling; if confidence in the ability to ensure the model's compliance with safety standards diminishes below an acceptable threshold, OpenAI is prepared to halt the expansion of its capabilities.
The launch of GPT-6 Astra marks a pivotal development in the AI landscape, potentially reshaping competitive dynamics as companies race to advance their own models and capabilities in pursuit of AGI.
Informational material. 18+.
