Jev reaches 35% of Vercel paid teams in two weeks

By VK

Published 18 Oct 2026, 09:48 UTC

Reporting from TechCrunch

TypeSafe AI's Jev decision model reached 35% of paid Vercel AI Gateway teams in two weeks, the fastest uptake Vercel has seen, at a tiny share of spend.

What it means for founders

Cheap calls make constant checks affordable. If TypeSafe's pricing holds, a decision model can sit on every step of a workflow: tagging tickets, routing requests, or vetting an agent's next action before it runs. That last use matters as agents misbehaving in the wild becomes a real operational problem, and sending every such check to a frontier LLM gets expensive fast.
Put it behind your own interface. Vercel says nearly a third of teams switched their primary lab in September. Wrap decision calls in a thin layer so you can swap Jev for a rival or an LLM fallback without rewriting product logic.
Budget for prices to move. Jev is still in early access. Track per call costs from day one and leave margin in case rates change.
Watch Vercel's next monthly index. It will show whether Jev keeps its reach and week two return rate once the novelty fades and the clones mature.

The story

Vercel released its monthly AI Gateway Production Index on Thursday, October 15, and the headline result belonged to a newcomer. Jev, a decision model from startup TypeSafe AI, was used by 35.3% of paid teams on the gateway within 14 days of going live there, the quickest uptake Vercel says it has ever recorded. Claude Sonnet 5.5 placed second at 18.1%. TechCrunch reported the numbers a day later; the full scorecard is in Vercel's report.

What Jev is and how it differs from an LLM

Jev does not write text. A developer passes it some input along with a fixed menu of allowed answers, and it returns a probability for each option. TypeSafe pitches this as a building block for automation: routing, classifying, scoring and guarding other systems, not chatting. The company says Jev runs on a transformer yet is not an LLM, and that fixing outputs in advance means it cannot return a malformed answer.

TypeSafe's launch post lists input pricing of $0.042 per million tokens, with outputs unbilled, and responses in 70 to 500 milliseconds. Vercel's data shows the same thrift from another angle: Jev handled more requests than any other model in September, yet it took only 0.27% of gateway spend across its first two weeks. Vercel itself moved its harmful command screening from GPT-6 Luna to Jev, according to TechCrunch.

Loyalty was solid: 55.6% of first week users came back in week two, fifth among September launches.

Who makes Jev

TypeSafe AI was started in 2024 by Diogo Almeida, a former OpenAI researcher, together with Sasha Sheng, previously a research engineer at Meta, and engineer Erik Gafni. Jev opened in early access on Tuesday, September 15. On October 9 the company announced a raise of $870 million, valuing it at $7.5 billion, in a round led by Andreessen Horowitz with Sequoia and DCVC. TypeSafe also claims that one in three Fortune 500 firms already use Jev.

OpenAI and Amazon have since shipped competing decision offerings of their own, including the Decisions API OpenAI unveiled at Dev Day.

What we don't know yet

How representative Vercel's view is, since labs with their own APIs, such as OpenAI and Anthropic, are likely undercounted on a third party gateway.
How well Jev's accuracy and confidence scores hold up on production work beyond TypeSafe's own evaluations.
Whether the pricing lasts. TypeSafe concedes it cannot prove its rates are not subsidized.
How Jev stacks up against the OpenAI and Amazon alternatives once developers test them at scale.