OpenAI says approximately 10,000 agents produced a proposed solution to the Navier–Stokes Millennium Prize Problem in 88 hours, followed by a 17-hour GPT‑6 Astra verification run. The proof still requires independent mathematical scrutiny, while competing researchers have raised unresolved questions about research provenance.
What changed
Multi-agent orchestration is being applied to open-ended discovery, not merely parallel software tasks.
Why it matters
At this scale, evidence lineage, duplicated work and evaluator independence become core infrastructure concerns.
Practical takeaway
Separate proposer and verifier pools; persist every hypothesis, dependency, rejection and source reference in an immutable experiment graph.
Analysis
A model’s verification is supporting evidence—not independent validation—when the verifier comes from the same model family and organisation.
Published
8 September 2026
