A small, year-old startup has produced the fastest-adopted model in the history of Vercel's AI Gateway, a marketplace that routes AI traffic for thousands of software teams. TypeSafe AI's model, called Jev, reached nearly 13% of all paid teams on the gateway within 24 hours of its mid-September launch, according to Vercel's own announcement — about twice the adoption of OpenAI's GPT-5.6 family and more than six times that of Anthropic's Fable 5.1 over the same window.
Jev is not a chatbot. It takes in program context and a set of questions, then returns typed answers — a choice from a list, a numeric score, or a yes-or-no call — each attached to a probability. Vercel says those outputs suit tasks that sit inside software pipelines rather than in front of users: picking which tool an AI agent should call next, deciding whether a workflow should continue or escalate to a human, or flagging outputs that need review before they reach a customer.
A Narrower Model, a Faster Build
TypeSafe AI emerged from stealth in mid-September with $40 million in seed funding led by DCVC, which valued the company at $200 million, according to SiliconANGLE. The company was founded in 2024 by chief executive Diogo Almeida, who worked on reinforcement learning from human feedback, InstructGPT, ChatGPT and GPT-4 during his time at OpenAI, along with co-founders Erik Gafni and Sasha Sheng. Almeida has framed the company's pitch as a correction to an industry obsessed with chat: "We've been optimizing for humans, and we're superhuman at pleasing humans," he said, arguing that software doesn't need a model that pleases people, just one that answers reliably.
TypeSafe calls Jev the first of a line it labels "System One" models, trained with a method it calls Reinforcement Learning for Calibrated Decisions. The company says the model responds in under 100 milliseconds and claims a cost of $0.39 per 1,000 workflow calls, well below the per-call cost of general-purpose language models — figures that are the company's own and have not been independently verified. Within three days of launch, according to the company, Cloudflare, LangChain and Langfuse had each integrated Jev into their own developer tools.
Early Lead, Unproven Durability
DCVC general partner James Hardiman said TypeSafe is tackling "one of the biggest remaining challenges in AI" by making models reliable enough to embed directly into products at scale, rather than leaving every decision to a general-purpose chat model. Vercel's own writeup is more measured about what the launch numbers actually prove, noting plainly that first-day adoption was "unmatched among recent launches" but that the real test is "whether that early adoption lasts."
That caveat matters, because Vercel's AI Gateway numbers measure how many paid teams tried a newly available model in its opening days, not how many kept using it, or what they used it for once the novelty wore off. Suggested use cases so far include insurance underwriting, invoice review, security-alert triage and acting as a guardrail layer alongside the larger language models that still do the writing and reasoning. TypeSafe is currently granting access through a waitlist rather than opening Jev to all developers at once, a rollout pace that will determine how much of its early lead on Vercel's platform translates into sustained use.