OpenAI Says GPT-6 Astra Scores Higher and Works Faster on Ironclad Contract Tasks
AI Agents & Automation
OpenAI Says GPT-6 Astra Scores Higher and Works Faster on Ironclad Contract Tasks
OpenAIOct 6, 2026
OpenAI says GPT-6 Astra scored higher and completed attempts faster than GPT-5.6 Sol across 11 research tasks developed with Ironclad.
OpenAI said on October 6, 2026, that it worked with contract-management software company Ironclad to train and evaluate agents on complex legal and business workflows. The collaboration produced 11 research tasks spanning legal, commercial, and procurement work, including configuring nondisclosure agreements, setting up approval processes, and adapting reusable clauses to a selected jurisdiction.
OpenAI reports that GPT-6 Astra achieved a mean rubric score of 55.0% across the tasks, compared with 41.6% for GPT-5.6 Sol. The company also estimates average time per attempt at 19.2 minutes for Astra and 37.0 minutes for Sol. These are research results from the limited task set, not measured customer time savings: OpenAI says the time figures are simulated estimates based on assumed model processing and generation speeds.
The tasks were assessed against 8 to 50 criteria each, depending on complexity. Ironclad provided hosted environments where models could practice, while researchers used reinforcement learning and synthetic tasks to improve performance. OpenAI says those synthetic tasks were built from publicly available contracts in the SEC’s EDGAR database after filtering intended to remove personal information; it says no OpenAI customer data, internal OpenAI contracts, or nonpublic Ironclad customer contracts were used for training or evaluation.
An internal model used during Astra’s development scored 63.7% on the same evaluation, according to OpenAI. The company says it aims to bring those further gains to future models. The results do not establish performance across all Ironclad workflows: they cover only the 11 research tasks, and OpenAI and Ironclad stress that human oversight remains important for workflows where business rules and exceptions must be preserved.
Industry Impact
AI advances in contract workflows
Legal operations, commercial contracting, and procurement
Overview
OpenAI reports that GPT-6 Astra scored 55.0% across 11 Ironclad-developed research tasks, versus 41.6% for GPT-5.6 Sol, and had a lower estimated time per attempt. For legal and business teams, the results suggest potential for agents to assist with bounded tasks such as configuring agreements and approval processes. However, the benchmark is limited, and the time figures are simulated rather than measured customer savings. Product and technology leaders should treat the findings as an early signal, not evidence of broad workflow readiness; human oversight remains important.
Key Impact Points
Astra improved mean rubric scores by 13.4 percentage points over Sol on the 11-task evaluation.
Estimated attempt time was 19.2 minutes for Astra versus 37.0 minutes for Sol, but these are simulated estimates, not customer time savings.
Results are limited to the research tasks; OpenAI and Ironclad say human oversight remains important for business rules and exceptions.
Opportunities
Scoped contract-workflow automation
Legal operations, commercial contracting, and procurement
The evaluation points to a potential opportunity for contract-management providers and enterprise teams to test AI agents on specific, rubric-defined tasks, such as configuring nondisclosure agreements, setting up approvals, or adapting reusable clauses to a jurisdiction. The reported results are research findings, not proof of customer savings or broad workflow coverage, so any product effort should begin with controlled pilots and human review. Teams could measure task accuracy, exception handling, review effort, and completion time using their own rules and data. OpenAI says its training and evaluation did not use customer contracts, which also makes customer-specific validation important before operational use.
Pilot agents for bounded contract tasks
Problem: Contract configuration and related business workflows involve detailed criteria, rules, and exceptions.
Solution: Pilot AI agents on narrowly scoped tasks such as NDA setup, approval configuration, or clause adaptation, with human review and task-specific evaluation.
Target Users: Legal operations, procurement, and contract-management teams.
Business Value: Assess whether agents can improve task completion quality or reduce effort in defined workflows; the article does not establish customer savings.
Revenue Potential: Potential for enterprise workflow or contract-management features if customer pilots demonstrate reliable value.
