OpenAI model claims 3D Navier-Stokes finite-time blowup proof, $15M to reproduce
johnseach · x · 2026-09-09
OpenAI says an internal model "significantly more capable than GPT-6 Astra" proved that 3D Navier–Stokes solutions can blow up in finite time — smooth start, smooth forcing, finite energy, then a vortex reaching infinite speed. That's one of the $15M Millennium Prize Problems.
Key facts and caveats:
- Reproduction cost: OpenAI told reporters reproducing the search as a customer would cost roughly $15 million
- Scale: 10,000 concurrent agents, 88 hours of search plus formalization, hundreds of billions of tokens, millions of agent messages
- GPT-6 Astra itself prices around $10/$50 per million tokens; the model that actually finished the proof sits above Astra, was trained during the run, and isn't public
- The post argues for two stacked gaps: closed weights, and a method (agent swarms + formal checkers + days of wall-clock) priced for frontier labs, not PhD students or research groups
- Takeaway: a landmark mathematical result, but a discovery engine behind an eight-figure invoice is not yet a scientific instrument for most researchers — inference and orchestration must get orders of magnitude cheaper
More from Models
- Founder praises GPT Astra for faster reasoning and sharper interpersonal guidance — morganlai · 2026-09-09
- DeepSeek V4.1 Flash Vision Beta tested in 5 visual tasks: far better and cheaper than V4 — cheezeerd · 2026-09-09
- WaveNet turns 10: long-context autoregression a year before Transformers existed — teropa · 2026-09-09
- OpenAI Codex goes down, disrupting developer workflows — cyrus_zei · 2026-09-09
- Fchollet: Early expert verdict on AI-generated math proof candidate is "no" — fchollet · 2026-09-09
- Codex user: one project burned 70% of the 20x weekly limit in a week — jdjohnson · 2026-09-09