A primary verified detail is the pricing structure: OpenAI has set the cost at $10 per million input tokens—the basic units of data the model processes—and $50 per million output tokens, matching the rates of its closest competitor, Anthropic.
While the pricing makes Astra one of the most expensive options on the market, OpenAI argues the model is more efficient, requiring fewer total tokens to complete complex tasks than its predecessors.
This release targets high-level professional work, such as software engineering and scientific research, where it reportedly outperforms rival models on technical benchmarks.
By positioning Astra for "computer use"—the ability to navigate software interfaces to perform multi-step actions like financial modeling or circuit design—OpenAI is attempting to cement its technology as essential infrastructure for enterprise-level automation.
To address safety concerns following a previous data breach, OpenAI has implemented stricter safeguards, including a refusal to perform advanced cybersecurity tasks such as discovering software exploits.
However, the company’s chief scientist noted that as models become more intelligent, "monitorability"—the ability for humans to track and understand the model's internal reasoning—is becoming more difficult.
Consequently, while Astra is designed to be more "aligned," or better at following human instructions and boundaries, OpenAI is prioritizing research into new ways to supervise these increasingly complex systems as they are integrated into professional workflows.