$2,000 Solved a Decade-Old Problem, But Startups Shouldn't Just Focus on Models

Just saw news from IT Home: OpenAI used an internal version of the Astra model, spending $2,000 in token costs, to solve 10 mathematical problems that had remained unsolved for at least ten years in one go. The math community blew up; Alex Kontorovich left two exclamation marks directly. Honestly, this hits harder than any Scaling Law demo—it proves inference costs can drop near zero while the capability ceiling has been pierced.