OpenAI Details GPT-6 Astra With Ironclad and Jump

OpenAI says GPT-6 Astra beat GPT-5.6 Sol on Ironclad contracting tasks and Jump Trading uses it for quant research.

On October 6, 2026, OpenAI published two separate accounts of GPT-6 Astra: a computer-use training partnership with contracting vendor Ironclad and a customer case study with Jump Trading. For Ironclad, staff and OpenAI users defined 11 legal, commercial, and procurement tasks judged against 8 to 50 criteria, with synthetic tasks and reinforcement learning used for practice. On that evaluation GPT-6 Astra averaged 55.0% versus 41.6% for GPT-5.6 Sol, which OpenAI called a 32% higher score, while estimated time per attempt fell from 37.0 to 19.2 minutes; an internal development model scored 63.7%. OpenAI said it is inviting a small number of other software companies to contribute similar professional tasks and secure test environments. Separately, Lucas Baker, head of LLM R&D at Jump Trading, said GPT-6 Astra supports multi-day coding and hypothesis-validation workflows with less human intervention than earlier systems, while human review, scoped signals, and monitored environments remain because finance is regulated. The reports do not contradict each other: one is a training and evaluation partnership, and the Jump piece is explicitly a case study rather than a new model launch.