OpenAI has achieved a mathematical proof for a 50-year-old unsolved problem using a publicly accessible model rather than experimental research-only AI, according to Ethan Mollick. The distinction matters: while previous major mathematical breakthroughs came from specialized research models, OpenAI this time deployed GPT-5.6 Sol Ultra with 64 parallel subagents that solved the problem in just under one hour.
The essentials
- Model: GPT-5.6 Sol Ultra (publicly available, not experimental)
- Method: 64 parallel subagents
- Time to solution: Just under one hour
- Significance: Genuine mathematical proof for a classic unsolved problem
What this signals
Historically, major AI-driven mathematical breakthroughs have been the result of specialized developments – models available only to research teams. Mollick emphasizes that OpenAI took a different path this time: the model is public. Anyone can use it. This suggests the boundary between research and production models is blurring. A public model capable of solving genuine mathematical problems is not just a benchmark win – it's a tool.
The use of 64 subagents is noteworthy: the model wasn't simply unleashed on the problem. Instead, a multi-agent architecture was employed, with different "thinkers" working in parallel, comparing and refining results. This mirrors actual mathematical collaboration more closely than raw compute.
Why this matters in practice
Mathematical proofs aren't merely academic curiosities – they have real applications in cryptography, optimization, materials science, and beyond. If a public AI model can solve such problems, it means: companies and research groups don't need to wait for proprietary systems. They can experiment today.
Yet questions remain. How reproducible is the result? How robust is the proof? Most importantly: how does this scale to even harder problems? Mollick doesn't provide these details – this is an announcement, not full documentation.
What this means for enterprises and research
The signal is unmistakable: public AI models are becoming competitive for specialized tasks. Organizations in pharma, materials science, financial modeling, or engineering should take note – not as hype, but as a practical tool. Equally important: achieving AI breakthroughs doesn't require owning a mega-model. Smart architecture (here: the 64 subagents) can be just as decisive. That's an opportunity for specialized teams, not just the big labs.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
