OpenAI said on October 6 that it was working with Ironclad to train and evaluate AI agents on contracting workflows across legal, commercial, and procurement work.

In OpenAI’s 11-task research evaluation, GPT-6 Astra averaged 55.0%, versus 41.6% for GPT-5.6 Sol, using each model’s best-performing reasoning setting.

The estimated time per attempt fell to 19.2 minutes from 37.0 minutes. OpenAI said the results did not cover all Ironclad workflows.

OpenAI said the timing figures were simulated estimates, not measured customer time savings.

It added that the training and evaluation used synthetic tasks based on public contracts from the US Securities and Exchange Commission’s EDGAR database, not OpenAI customer data, OpenAI’s internal contracts, or nonpublic Ironclad customer data or contracts.

Ironclad, a US-based contract lifecycle management software company, lists an OpenAI integration, and OpenAI features the company in a contract review case study.

PwC also discusses the role of AI agents in contract lifecycle management.

🔗 Source: OpenAI

Recent OpenAI developments