Jev's surge is adoption, not pricing power: TypeSafe priced it at $0.042 per million input tokens (company docs, Sep 15), and Together's $17 copy matched that. Vercel says nearly 13% of paying teams used Jev within 24 hours (Vercel blog, Sep 18); under a free promotion it took 26.7% of requests but 2.0% of tokens on Sep 24. The reported valuation of over $10B (The Information and FT, citing people familiar; unsigned) buys an option on the default-judge seat, which distributors hold.
Investors: do not enter at $10B; back the harnesses that hold the default seat.
Founders: trial routing on Jev-class models, never as a security boundary.
Falsifier: per Vercel's dataset=models export, any 7-day average of Jev's request share below 13% from Sep 26 to Oct 31, 2026 verifies the call; at or above 20% marks it wrong.
Ledger update (Sep 29): Sep 26 to 28 averaged 23.1%; no full window yet, so pending, wrong-condition holding.
Ledger update (Oct 2): Jev's request share averaged 23.1% over the first six complete post-promo days, so the wrong-call condition holds for the Oct 31 ruling; five rivals shipped decision models from Sep 29 to Oct 1, but neither OpenRouter nor Vercel lists one at or below $0.02/M, so falsifier 2 is not triggered.
Disclosure: written with Claude tools; Claude Fable 5.1 is TypeSafe's pricing comparator; FutureX holds Hugging Face, host of the open copies; the author is a Vercel customer; FutureX holds no position in TypeSafe, DCVC, Together, OpenAI, Cloudflare, LangChain or ConvAI.
This diary does not constitute promotion of, or an offer to subscribe for, any fund product.