Advancing computer use with Ironclad

OpenAI announced a partnership with AI contracting leader Ironclad to train and evaluate AI agents on complex contracting workflows, aiming to advance computer use for professional work. Together, the companies identified 11 challenging tasks across legal, commercial, and procurement operations, such as configuring agreements, approvals, and reusable legal terms. Ironclad provided hosted software environments where models could practice these workflows, allowing researchers to use reinforcement learning and detailed evaluation criteria ranging from 8 to 50 metrics per task. The collaboration resulted in GPT-6 Astra, the first frontier model trained on Ironclad tasks. In research evaluations across the 11 tasks, Astra achieved an average score of 55.0 percent compared to 41.6 percent for GPT-5.6 Sol, while estimated average time per attempt fell from 37.0 minutes to 19.2 minutes. An internal model used during Astra's development scored even higher at 63.7 percent. OpenAI noted that human oversight remains essential for contracting tasks due to the need to handle exceptions while preserving business rules. Following this partnership, OpenAI is inviting additional software companies with domain expertise, secure testing environments, and concrete failure examples to collaborate on solving complex professional tasks that current AI agents cannot yet complete reliably.