OpenAI evaluated a computer-use model on 11 tasks based on Ironclad’s actual contract workflows. Published results gave GPT-6 Astra a criteria-satisfaction score of 55.0%, above the comparison model’s 41.6%, with lower estimated task times as well. The score aggregates multiple detailed criteria rather than representing the success rate for entire contract workflows. An internal model evaluation and estimated times do not establish that legal work can be handled without people.