This picture taken on January 23, 2023 in Toulouse, southwestern France, shows screens displaying the logos of OpenAI and ChatGPT. - ChatGPT is a conversational artificial intelligence software application developed by OpenAI. (Photo by Lionel BONAVENTURE / AFP) (Photo by LIONEL BONAVENTURE/AFP via Getty Images)AFP via Getty ImagesThe cost of producing new results on ten long‑standing mathematical problems just fell to $2,000, according to OpenAI, which says its Astra model generated machine‑checkable proofs for questions that had resisted human progress for decades.OpenAI published the work on August 1 and used it to give its next major model family a name: Astra. The results run across group theory, high-dimensional geometry, coding theory, quantum complexity, lattice cryptography and extremal combinatorics. They arrived as a 249-page manuscript collection and, alongside it, something the field has not seen attached to an AI claim before at this scale: a machine-checkable certificate for every single result.The problems were not textbook exercises dressed up as discoveries. Each had been open for at least ten years, most of them far longer, and several sit at the center of their subfields.A construction establishing the existence of non-sofic groups, a question that has occupied group theorists for years.A disproof of Connes's rigidity conjecture, a long-standing problem in the theory of von Neumann algebras.An improvement to the general upper bound on sphere-packing density in high dimensions, a bound that had stood since 1978.Three problems come from the catalogue of open questions left behind by Paul Erdős.The announcement follows another result from May, when the same model family reportedly disproved the Erdős unit distance conjecture, an 80-year-old problem in discrete geometry that had resisted every serious attempt since 1946.Fields Medalist Tim Gowers said he would have recommended the proof for publication in a top mathematics journal without hesitation. A team of nine mathematicians, including Gowers and Noga Alon, later published a companion paper explaining the proof in a way that human mathematicians could more easily follow.Thomas Bloom, who maintains the Erdős problem catalogue, called the August results “big news” and said they were even more significant than the earlier unit distance result. OpenAI researcher Noam Brown added a note of perspective: “Sadly, no Millennium Prize Problems (yet).”Why These Proofs MatterEvery AI capability announcement of the past three years has shared one weakness. The company making the claim is also the only party able to evaluate it. Benchmarks get contaminated, demos get curated and the outside world is left arguing about whether the number means anything.Astra's results were formalized in Lean, a proof assistant that verifies mathematical arguments step by step, and the certificate files were published on GitHub under an open license. Anyone can download them and run the checker. If a single step in the argument fails to follow from the previous one, the software rejects it. There is no interpretation involved and no committee to convince.Ordinarily a claimed proof of a famous problem enters peer review, where human referees spend months working through the logic and the field waits to hear whether it held up. The Lean certificate collapses that timeline to the length of a download. Verification and publication arrived on the same day.That property is rare, and it is the reason this announcement reads differently from a benchmark score. A verified proof does not require the reader to believe the entity that produced it.What Skeptics Get RightThree objections are worth taking seriously, and the honest reading concedes all three.The problem set may have been selected. OpenAI chose which results to publish, and the $2,000 figure covers the successful runs rather than every attempt the model made, which makes it a cost of publication rather than a cost of discovery. Outside researchers have also noted that OpenAI staff helped prepare the papers and formalize the arguments, with the company saying the mathematical content came from Astra. And nobody outside the company can run the model that did the work, so the result cannot be reproduced independently, only checked.Gary Marcus, a reliable critic of the field, called the release amazing but vastly oversold. Some specialists expect that once the dust settles, a few of the ten will look genuinely surprising and the rest will be classified as reachable problems that nobody had gotten around to attacking.Grant every objection and the structural point survives intact. Whether the ten were cherry-picked or not, the certificates still verify. A curated result that mechanically checks is a different object from a curated result that does not.Where Verified Output Already WorksDeployment stalls for a reason that has little to do with how clever the model is. A business cannot use output it has no way to check, and human review stops scaling long before the output does. An analyst can read a paragraph carefully. Nobody reads ten thousand of them, which is where a great deal of enterprise AI work runs aground.A handful of industries never had that problem, because they built machine-checking into their workflow decades ago. Chip design is the clearest case. Formal verification tools mathematically prove that a circuit does what its specification says, and that infrastructure existed long before anyone thought to point a language model at it.The consequence is visible in what those companies are shipping. Cadence announced at Computex in May that it had extended its design agent to full autonomy. The agent runs hundreds of simulations against the company's Jasper formal verification engine, compressing a five-week validation loop to under a day. Synopsys sells the same category of tool in VC Formal, which uses static analysis to prove a design correct rather than testing it case by case.That speed has less to do with courage than with arithmetic. A chip designer can check the machine's answer automatically and cheaply, so a wrong answer costs almost nothing to catch.The same structure holds anywhere a proof obligation already exists. Cryptography, safety-critical software, and hardware verification all share it. The companies that own those checking layers are the ones whose products get more valuable as AI output gets more voluminous, because volume is exactly what makes human review break down.The cost of producing a hard answer just fell to almost nothing, and the constraint moved to proving the answer is right.