Jev, TypeSafe's decision model, is Vercel's fastest-adopted ever
A non-chat model that returns a boolean, score or choice for cents — wired into Vercel, Cloudflare and LangChain within days as open weights take the gateway.
Copy markdown
Typed answers, not chat
Jev, from ex-OpenAI researcher Diogo Almeida's TypeSafe AI, is a "decision model": instead of generating text it returns one typed result — a boolean, a score, or a choice from a fixed list — for the high-volume calls agents today waste on frontier chat models, like intent classification, tool selection and risk scoring. TypeSafe trained it with a method it calls Reinforcement Learning for Calibrated Decisions.
Fastest launch in AI Gateway history
Vercel says typesafe/jev hit ~13% of paid teams in its first 24 hours — more than double any prior model launch, including the GPT-5.6 family — and TechCrunch now puts it near 35% of Vercel teams. Vercel, Cloudflare, LangChain and Langfuse all wired it in within three days.
$0.042/M in, free output, 70-500ms
That's the real unlock: route your classifiers and risk checks to Jev and the per-call cost of those decisions collapses versus a chat model priced for open-ended generation. Vendor benchmarks (flagged as vendor numbers) claim up to ~445x cheaper and ~194x faster on narrow decision tasks. Weekend project: move one router or intent classifier in your agent off a chat model and onto a decision model.
Open weights now move most gateway tokens
Jev's rise fits a bigger shift: open-weight models now handle about 71.5% of Vercel AI Gateway token volume but only ~23% of the spend — cheap, specialized models eat the volume while a few frontier models still take the dollars. The takeaway: build model-agnostic so you can route each call to the cheapest thing that clears the bar.
Copilot auto-fixes code-scanning alerts
GitHub moved agentic autofix for code scanning alerts to general availability: Copilot now proposes fixes for security findings straight from code scanning, out of preview. If you run GitHub Advanced Security, your vulnerability backlog gets a drop-in agent that drafts the patch for review.