Jev 24 小时覆盖 Vercel 近 13% 付费团队
Jev 靠「只做决策不写句子」在 Vercel AI Gateway 快速铺开，返回带概率的类型化答案，已被用于 agent 路由、护栏、预测表格和实时游戏。
The fastest-adopted AI model in Vercel AI Gateway history literally can't write a sentence. Jev reached nearly 13% of Vercel's paid teams in 24 hours. And the reason is surprisingly simple: Most software doesn't need an AI to write 500 words. It needs AI to make a decision. Which agent should run? Approve or escalate? Retry or stop? Which tool should I call? Is this risky? Which option is best? Jev takes the current state + predefined choices and returns typed answers with probabilities. No paragraph to generate. No JSON to parse. No waiting for 200 tokens to stream. TypeSafe prices it at $0.042 per 1M input tokens, with typical responses around 70–500ms. So developers are putting it inside loops where calling GPT or Claude constantly would be too slow or expensive. People are already using it for agent routing, guardrails, predictive spreadsheets, and real-time games. One demo even has Jev controlling all 4 characters in Smash Bros against itself. The interesting part isn't the game. It's what happens when AI decisions become cheap enough to put almost everywhere in software. https://x.com/maubaron/status/2100738237237002706/video/1