[AINews] TypeSafe/Jev at >$100M ARR, $7.5B valuation 3 weeks after launch

TypeSafe is now in the middle of the loudest new category fight in AI: decision models. The company announced its “Series AI,” and Sequoia reportedly leaked that it crossed $100M ARR in its first week. That alone would be wild. But it landed in a week when almost everyone seems to have shipped some version of the same thing.

These models don’t write essays or chat much. They return typed answers: a probability, a pick from a list, a score. One forward pass, no free-form ramble. Jev has become the reference point, and the rush around it has been fast enough that even the recap notes call out how quickly the format spread.

The list of entrants is already crowded. OpenAI’s Decisions API handles three request types and runs on GPT-6 Luna. Microsoft has Decision-1. Perplexity says its pplx-decider-v1.1-27b leads the Decision Bench with 94.5% across 1,071 cases. Cloudflare’s clef now takes audio, video, image and text, and its clef-flash is cheaper than Jev. Vercel picked up Liquid d1 for classify, route and score tasks. LangSmith is using Jev as a judge inside traces.

And the pitch is pretty simple: a lot of agent work is not creative writing, it’s yes/no calls. That’s why harnesses care. LangChain says routing each task to the cheapest adequate model cut median Open SWE cost per task by 64%. Apple and CMU’s SSR paper pushes the same idea into agents, and the reported gains are real but mixed: per-turn reasoning latency drops by more than 90%, while end-to-end question latency falls 28–54%.

So yes, everyone is cloning the Jev API. The interesting part is not whether the copies exist. It’s that typed decisions have gone from niche infrastructure idea to a product category in one very loud week.