TypeSafe AI's Decision Model Jev Becomes Vercel's Fastest Adopted Launch

TypeSafe AI's Jev model became the fastest-adopted model in Vercel AI Gateway's history within three days of its September 15, 2026 early-access launch, reaching nearly 13% of paid teams by hour 24 — more than twice the adoption rate of any prior Vercel launch, including the GPT-5.6 family. Jev, built by CEO Diogo Almeida's startup on a training method called Reinforcement Learning for Calibrated Decisions (RLCD), answers typed yes/no, multiple-choice, or scored questions in 70 to 500 milliseconds with a calibrated probability, priced at $0.042 per million input tokens with free output. Vercel CEO Guillermo Rauch said Jev ran up to 18 times faster than GPT-class models at the 95th percentile and was more accurate on the same tasks, while Forbes reported real-world cost cuts closer to 100x.

TypeSafe AI's new model Jev skips chat entirely and just answers yes, no, or a number, and it became the fastest-adopted model in Vercel AI Gateway's history within three days of launch. Diogo Almeida helped build ChatGPT at OpenAI. His new startup just shipped a model that refuses to chat. TypeSafe AI, the company Almeida founded after leaving OpenAI with a $40 million seed round, opened early access to Jev on September 15, 2026. Within three days, Vercel's AI Gateway, Cloudflare Workers AI, LangChain, and Langfuse had all wired it in. Vercel says it's the fastest-adopted model in the Gateway's history: by hour 24, Jev was running for nearly 13% of paid teams, more than twice the adoption rate of any model Vercel had launched before, including the GPT-5.6 family. So what does it actually do? Jev doesn't write essays or draft emails. You send it a chunk of state, a support ticket, a tool call, a user request, along with a typed question: a yes or no, a choice from a list, a score on a scale. It answers in under half a second with a calibrated probability attached, then moves on. No chit chat. No formatting. No wasted tokens explaining its reasoning. TypeSafe built Jev on what it calls Reinforcement Learning for Calibrated Decisions, or RLCD, a training approach aimed at getting probabilities right rather than prose fluent. Input costs $0.042 per million tokens. Output is free, because there isn't much of it: a boolean, a score, a label. Response times run 70 to 500 milliseconds. TypeSafe's own benchmarks put Jev at up to 194 times faster and 445 times cheaper than comparable frontier language models on these narrow decision tasks. Ilya Sutskever's SSI Has Raised $8 Billion and Shipped Nothing At All https://startupfortune.com/ilya-sutskevers-ssi-has-raised-8-billion-and-shipped-nothing-at-all/ Ilya Sutskever's Safe Superintelligence has raised about $8 billion, including a fresh $5 billion from Nvidia, without shipping a single product. Sutskever says the industry's scaling era is over and SSI is betting on something else entirely, but two years in, the company still hasn't shown the world what that something is. - how to price a SaaS product for enterprise https://startupfortune.com/ilya-sutskevers-ssi-has-raised-8-billion-and-shipped-nothing-at-all/ - cold email template that gets replies from investors https://startupfortune.com/ilya-sutskevers-ssi-has-raised-8-billion-and-shipped-nothing-at-all/ Vercel didn't just take TypeSafe's word for it. According to Vercel CEO Guillermo Rauch, Jev came in up to 18 times faster than GPT-class models at the 95th percentile, and more accurate on the same tasks. That's a smaller multiple than TypeSafe's own marketing numbers. Worth flagging. But it still confirmed the direction: Jev wins, and wins by a lot, on the boring calls agents make constantly. Forbes reported the real-world cost cut lands closer to 100x once you account for how these decision calls actually get used in production. That's the number that counts. The bill hiding under the hood TypeSafe frames this as System One versus System Two thinking, borrowing Daniel Kahneman's split between fast instinct and slow deliberation. An AI agent doesn't need a frontier model to decide which of five tools to call, whether an output looks like a jailbreak attempt, or how to route a support ticket. Those are System One calls: cheap and low-stakes individually, expensive in aggregate. A single customer support agent might make a dozen classification and routing decisions before it ever generates one sentence a human actually reads, and every one of those calls, run on a general-purpose model, costs real money and adds real latency. That's the bill nobody talks about. Frontier-model pricing gets debated constantly. The routing and classification calls sitting underneath every agent, invisible on a single receipt, rarely do. The adoption speed is the part that should actually worry competitors. Model launches usually take weeks to work their way into the infrastructure agent builders touch every day. Jev went from early access to four major platforms, run by three different companies, in 72 hours. Vercel didn't wait for a slow rollout. Neither did Cloudflare, which shipped native Workers AI support routing Jev calls through its own edge network, with Vercel's Gateway kept only as a fallback path. For founders building agentic products, the pitch is straightforward. Every dollar spent routing, verifying, and classifying is a dollar not spent generating the text a user actually reads, and that dollar has stayed invisible because it rides inside a single frontier-model bill. Jev turns it into a line item you can shrink on its own, without touching the model that handles the parts of the product a customer actually sees. Whether that holds up once Jev moves past early access and into full-scale production traffic is the question TypeSafe hasn't answered yet. Also read: Sony Music and Universal Music Sue Suno Again Over Its New V6 Model https://startupfortune.com/sony-music-and-universal-music-sue-suno-again-over-its-new-v6-model/ • A Ukrainian Veteran Built a Mental Wellness App Around a Reset, Not Another Routine https://startupfortune.com/a-ukrainian-veteran-built-a-mental-wellness-app-around-a-reset-not-another-routine/ • Anthropic Hires Accenture to Sit Inside Its Walls and Hunt for Dangerous AI https://startupfortune.com/anthropic-hires-accenture-to-sit-inside-its-walls-and-hunt-for-dangerous-ai/ Sword Health Is Buying Headspace for Up to $300 Million in Cash https://startupfortune.com/sword-health-is-buying-headspace-for-up-to-300-million-in-cash/ Sword Health, a Portuguese-founded AI healthcare startup, is buying Headspace in an all-cash deal worth up to $300 million, a steep drop from the $3 billion valuation Headspace held in 2021. The deal folds Headspace's 100 million users and 15,000-plus clinical providers into Sword's broader AI Care platform, closing in Q4 2026. - AI healthcare startup acquires meditation app Headspace https://startupfortune.com/sword-health-is-buying-headspace-for-up-to-300-million-in-cash/ - valuation decline of wellness companies in 2026 https://startupfortune.com/sword-health-is-buying-headspace-for-up-to-300-million-in-cash/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.