Jev 45 秒跑完 586 页内链审计，花费 0.21 美元
作者用 Jev 做站内链接审计，45.1 秒处理 586 页、放置 584 条链接并拒绝 139 页硬塞，总成本 0.21 美元；同样任务 Claude Opus 5 只跑完 21 页就花了 1.43 美元。
We’ve been massively overpaying for AI marketing work that doesn’t need frontier intelligence. I ran Jev across 586 pages of a site for an internal linking audit. In 45.1 seconds, it processed every page, rebuilt the internal link map, placed 584 links, and refused to force links onto 139 pages where nothing actually fit. Total cost: $0.21. I gave Claude Opus 5 the same 586 pages, same queue, same rubric, same clock. It got through 21 pages and spent $1.43 before the run stopped. That works out to roughly 190x more cost per page. A full Opus pass at that rate would have cost around $43. The interesting part isn’t that Jev is “better at SEO.” Internal linking is simply the kind of marketing task where using a frontier model can be overkill. The job is basically thousands of small classification decisions: should page A link to page B, and does relevant anchor text already exist in the copy? We’ve spent the last few years asking which AI model is smartest. For marketing teams, the more useful question might be: what is the cheapest level of intelligence that can reliably make this specific decision? As AI gets embedded deeper into marketing workflows, model routing will matter as much as model capability. The best AI stack won’t run everything through the smartest model. It’ll know when intelligence is actually worth paying for.
这条是引用着另一条写的：x.com/wwardenn/status/2102842288619814925