OpenAI's Unreleased Astra Cracked Ten Math Problems Nobody Had Moved in a Decade — for About $2,000 in Compute
About This Episode
On August 1, OpenAI published ten new results in mathematics and theoretical computer science produced by an internal version of Astra, the model it calls its next major model. Each argument was formalized into a machine-checkable Lean certificate and released publicly. OpenAI says the tokens needed to find all ten would have cost roughly $2,000 at its Sol API rates.
Our Take
OpenAI just put a receipt on frontier research — about $2,000 for ten problems the field hadn't moved in a decade, with machine-checkable proofs attached — so the question stops being whether the machine is smarter than the expert and becomes whether it is cheaper and checkable, which is the same test that has already been applied further down the ladder.
Continue Reading on Unscarcity
The AI Talent Paradox: When Genius Is the Last Scarcity
The article's framework holds that when compute and capital go abundant, elite human genius becomes the bottleneck — and a $2,000 run on decade-old open problems is the first serious test of whether that bottleneck holds.
The Substitution Threshold: When You Stop Being the Cheapest Option
The threshold is defined by cost and reliability rather than intelligence, and the Lean certificates plus the price tag supply exactly those two accounting numbers for the most credentialed cognitive work there is.