Jev Maker TypeSafe AI Hits $7.5B Valuation Just 24 Days After Launch

On October 9, TypeSafe AI, maker of Jev, announced an $870 million Series A at a $7.5 billion post-money valuation — just 24 days after Jev's public launch. Funding swelled ~22x and valuation multiplied 37.5x. Bloomberg, TechCrunch and a16z's official blog all confirmed the deal: a16z, Sequoia and DCVC piled in, with a16z partner Martin Casado — widely regarded as the godfather of infrastructure investing — personally joining the board.

What Makes a Model That Can't Chat Worth $7.5B

Jev's counterintuitive bet: it doesn't chat or write code. It strips out text generation entirely and does exactly one thing — making decisions. For high-frequency structured calls like routing, classification, scoring and multiple choice, it returns probabilities instead of prose. Pricing is equally aggressive: output tokens are free forever, input costs $0.042 per million tokens. Co-founder and CEO Diogo Almeida worked at Google Brain and OpenAI, was a core author of the InstructGPT paper and contributed to RLHF, ChatGPT and GPT-4 — one of the people who taught AI to talk nicely, now building an AI that stays silent.

Machine Traffic Is What Excites Investors

Vercel reported that within 24 hours of Jev's arrival on its AI Gateway, nearly 13% of paid teams adopted it — the fastest adoption in Gateway history and twice any previous model launch. Unlike chatbots riding human curiosity, Jev's volume comes from backend software classifying, scoring and choosing around the clock. As one line quoted in Chinese coverage put it: "Humans sleep. Code doesn't."

"Humans sleep. Code doesn't." — TypeSafe AI CEO Diogo Almeida, via Latent Space interview (as reported by QbitAI)

The Big Tech Siege, in 24 Days

Once the category was validated, imitators rushed in: Microsoft unveiled Microsoft-Decision-1 (post-trained on Qwen3.5-9B) two hours after the funding announcement; OpenAI opened its Decisions API public beta on October 6, powered by GPT-6 Luna and claiming up to 10x the speed of the same model via Responses API; Cloudflare had already open-sourced the Jev-API-compatible Clef and Clef-flash on October 1. From no man's land to main battlefield took barely three weeks.

The NineZenith View

Jev's rise signals something simple: in enterprise workflows, the most valuable step is often not "generation" but "judgment" — content moderation, ticket triage, routing, risk scoring are all high-frequency, cheap, reliability-first decisions. NineZenith likewise uses a hybrid "large model for generation + lightweight decision models" architecture to compress costs. When designing AI workflows, isolate which steps only need a decision — that exercise alone often cuts the majority of a token budget.

(Compiled from public reports by Bloomberg, TechCrunch, a16z, Vercel, Pulse 2.0 and QbitAI)