OpenAI Reports Navier-Stokes Breakthrough, With GPT-6 Astra Used for Lean Verification

OpenAI disclosed a significant advancement in AI-assisted research on the Navier-Stokes Millennium Prize Problem, revealing that an internal model named GPT-6 Astra generated a mathematical proof demonstrating that three-dimensional incompressible Navier-Stokes dynamics can develop a finite-time singularity. The company utilized GPT-6 Astra to formalize the proof in Lean, a formal proof language, which allowed for machine-checkable verification.

OpenAI emphasized that the internal model significantly outperformed GPT-6 Astra in generating the proof, but the latter played a crucial role in formalizing the argument. The timeline for the research effort began around August 28, 2026, with the agent-based system running for approximately 88 hours, followed by an additional 17 hours for Lean formalization.

OpenAI has also acknowledged concurrent work by Tristan Buckmaster from New York University and Levent Alpöge of Anthropic, acknowledging priority in this field. While the result is a substantial claim in mathematical research, it should be noted that OpenAI has not yet announced accessibility or pricing for the research system, nor has it provided evidence for customers to reproduce the Navier-Stokes workflow.

The announcement highlights the potential role of AI in structuring workflow, with one system generating candidate solutions and another verifying structure, consistency, or compliance with explicit rules, even though Lean formalization is specialized and may not be directly applicable to everyday business use.