OpenAI teams up with Ironclad to train contract AI agents on real professional workflows
On October 6, OpenAI announced a research partnership with Ironclad that turns 11 contract workflow tasks into training and evaluation problems; on the eval GPT-6 Astra scores 32% higher than GPT-5.6 Sol and runs 48% faster per attempt.
On October 6, OpenAI announced a research partnership with AI contract management company Ironclad, framing complex contracting workflows as training and evaluation tasks to advance computer-use capabilities for real professional work. Ironclad is the first partner; together the teams designed 11 tasks spanning legal, business and procurement workflows, including setting up NDAs, configuring procurement approval flows and updating reusable legal clauses for the applicant's selected jurisdiction. Skilled users take roughly 30 to 40 minutes per task.
OpenAI says GPT-6 Astra is the first frontier model trained on Ironclad tasks. In the company's research evaluation, GPT-6 Astra met about 94 percent of task criteria in an estimated 20 minutes, while GPT-5.6 Sol met about 85 percent in an estimated 32 minutes. That is roughly a 32 percentage-point lift on average score and 48 percent less estimated time per attempt for GPT-6 Astra. The evaluation used simulated tasks built from contracts publicly available in the SEC EDGAR database after filters to remove personal information; no OpenAI customer data, OpenAI internal contracts or non-public Ironclad customer data was used for training or evaluation.
OpenAI frames the collaboration as targeting the complexity of real multi-step business workflows: when a legal operations team builds a software procurement flow, finance may require approval above a certain dollar amount, security may need to vet specific applications, legal may need to review non-standard clauses, and the builder has to translate those requirements into request intake forms, document templates, approval rules and final agreement records. The challenge is not just single-step correctness but ensuring the resulting workflow holds across every scenario it was designed to cover. Ironclad CTO Sunita Verma said working with OpenAI's researchers brings these real-world challenges directly into model development. OpenAI says it is inviting a small number of software companies with similarly deep domain expertise into similar collaborations.