Jev 40 秒拆解 8724 条广告
用 typesafe-sdk 让 Jev 对 724 条在投广告做结构化判断，输出钩子、CTA、认知阶段等字段，中位耗时 216ms、总成本 9 美分。
724 live ads. 40 seconds. 9 cents. pip install typesafe-sdk That's Jev, the new model from TypeSafe. It doesn't write a single word. It only makes decisions, and it makes them in milliseconds. The demo: 37 brands, 8,724 judgments, median 216 ms per ad. Here is what it pulled from every ad in one run: → the hook and the format → the offer and the CTA → which awareness stage it targets → where the landing page breaks the promise of the ad Under the hood you get typed answers with a probability for every option. No broken JSON, no invented fields. And here is the part that got me: On TypeSafe's own benchmark it runs 193x faster and 444x cheaper than frontier LLMs. Input is $0.042 per 1M tokens. Output is free. Early access, API key after the waitlist. SDKs for Python and JS, MIT. The ad teardown itself is coming to StealAds and MCP. Link in the comments