The model notably scored 98% on the FrontierMath Tier 4 benchmark and reached 99.9% on ARC-AGI-3, a test designed to measure general intelligence and reasoning.
The update introduces major advancements in "computer use," a technical capability that allows the AI to interact directly with a user’s desktop environment to manage multistep workflows, such as generating complex documents or spreadsheets in the background.
OpenAI reports that Astra is nearly twice as fast at these tasks than its predecessors, while also achieving a "Critical" threshold in cybersecurity capabilities.
For software engineers using the Codex platform, the model introduces a new context preservation mechanism that allows it to keep detailed notes across long sessions without the typical data loss associated with summarization.
Following the initial release to Pro subscribers and specific cybersecurity partners, OpenAI plans to expand access to ChatGPT Plus, Business, and Enterprise users, as well as developers using the OpenAI API and AWS.
While the model will eventually become the default for these tiers, Enterprise administrators must manually enable the feature at launch.
The company is implementing this phased rollout due to the model’s high level of sophistication and its potential impact on cybersecurity environments.