TypeSafe AI Launches Jev, a Decision-Focused Model Claimed to Be 10x Faster and One-Tenth the Cost of LLMs

Summary

San Francisco startup TypeSafe AI has released Jev, an AI model designed for binary and multiple-choice decisions rather than conversational text generation. Jev returns structured decision results and confidence scores that software can use directly. The company says software development often consists of many small decisions and claims Jev is 10 times faster than large language models for these tasks while costing one-tenth as much. In a company demonstration, Jev classified a batch of customer messages in 2.3 seconds for about $0.001, compared with roughly 20 seconds using a chat-oriented large language model. TypeSafe AI presents customer-ticket classification, agent tool selection, real-time X feed filtering, and live meeting-transcript processing as example uses. These tasks include identifying refund or escalation requests, selecting the right tool, distinguishing posts by type, and determining whether a speaker has finished or made an actionable request. Developers can decompose complex workflows into many yes-or-no or multiple-choice checks performed on each message, tool call, or utterance. Jev is currently in early trials and available through an API, including via the listed OpenRouter model page. TypeSafe AI positions it as a “System One” model for fast internal software decisions, not as a general chat model.