TechCrunch's widely-cited "under $50" figure for training s1 is attributed to the paper, but the paper contains no such number

s1distillationcostquantitativeverificationtier-3-journalismsource-attribution

TechCrunch's February 5, 2025 article, headlined "Researchers created an open rival to OpenAI's o1 'reasoning' model for under $50," opens: "AI researchers at Stanford and the University of Washington were able to train an AI 'reasoning' model for under $50 in cloud compute credits, according to a new research paper released last Friday." No dollar figure of any kind appears anywhere in the paper the sentence names as its source — see claim-s1-paper-has-no-dollar-figure-only-gpu-hours — so the attribution itself does not hold up against the document it cites. This is a claim about what TechCrunch says and what the paper says, not a claim about what s1 actually cost; the "under $50" figure is recorded here [unverified-quant — needs primary] per the sourcing floor for quantitative claims (Tier 1–2 required; no Tier 1–2 source states it).

The same article's only dollar figure attributed to a named human, rather than to the paper, is different from the headline: "Niklas Muennighoff, a Stanford researcher who worked on the project, told TechCrunch he could rent the necessary compute today for about $20." This is a present-tense rental estimate ("today"), not a record of what the s1 team spent, and it is a different number from the headline's "under $50" — a second, unresolved figure inside the same article, on the same claim, sourced only to a reporter's relay of a quote rather than Muennighoff's own venue or a transcript. No independent record of his $20 estimate, or the per-GPU-hour rate behind it, was found. The headline number and the one sourced dollar figure in the article's own text disagree, and neither resolves against the paper.