智能体
www.aibase.com ·
TypeSafe AI 推出决策专用模型 Jev：比大语言模型快 10 倍、成本仅十分之一
旧金山初创公司 TypeSafe AI 发布决策专用模型 Jev，不生成文本，而是直接输出可被程序使用的结构化判断结果（yes/no 或多选）并附带置信度。公司称在软件开发中的大量决策类任务上，Jev 比传统大语言模型快 10 倍、成本仅十分之一，并演示了客服工单智能分类、AI Agent 工具调用选择、X 信息流实时过滤、会议记录实时判断等场景。官方演示称一批消息打标仅需 2.3 秒、约 0.1 美分，而聊天式大模型约需 20 秒。Jev 目前处于早期试用阶段，通过 API 提供，被定位为软件内部快速结构化决策的“System One”模型。上述性能与成本数据均来自公司自述，尚无独立验证。
AI推荐理由 TypeSafe AI 发布面向结构化决策的专用模型 Jev，只输出 yes/no 或多选结果并附置信度，宣称在客服工单分类、Agent 工具选择、信息流过滤、会议记录实时判断等场景比通用大模型快 10 倍、成本约十分之一。文中给出 2.3 秒/0.1 美分对约 20 秒的对比，但均为公司自述演示，无第三方评测或技术细节；模型处于早期试用阶段，仅通过 API 提供，实际效果与泛化能力尚待验证。
62
正文
TypeSafe AI Launches Decision-Specific Model Jev: 10 Times Faster and 1/10th the Cost of Large Language Models
AIbase基地
Published inAI News · 4 min read · Sep 21, 2026
10
San Francisco startup TypeSafe AI recently launched a new AI model called Jev. Unlike traditional large language models, Jev is specifically designed to answer yes/no questions and multiple-choice questions, and provides confidence scores. The company claims that a large portion of work in software development involves a series of decisions, and Jev is 10 times faster and costs only one-tenth of traditional large language models for such tasks.
According to TypeSafe AI, Jev does not generate text but directly outputs structured decision results that can be used directly by programs. Here are several practical application scenarios demonstrated by the company:
More efficient intelligent classification of customer service tickets. The system can quickly determine whether a ticket involves cancellation, refund, angry user, bug report, or needs escalation. Official demonstrations show that Jev can tag a batch of messages in just 2.3 seconds, costing about 0.1 cents; the same task using a chat-based large language model takes about 20 seconds.
Accelerate AI agent tool calls. When an AI agent needs to choose the appropriate tool, Jev can accurately select the correct tool at a lower cost and in less time.
Real-time filtering of X (formerly Twitter) information streams. Each reply can be seen as a multiple-choice question: Is it a genuine question, a popular opinion, promotional content, emotional provocation, or mere mischief? Jev can perform real-time filtering while users scroll, with extremely low costs.
Real-time processing of meeting records. For conversation content, Jev can instantly determine: Is it addressing the assistant? Has the other person finished speaking? Is a specific request being made? Are there any actions needed?
The company emphasizes that developers can break complex problems into a large number of small yes/no or multiple-choice decisions, which Jev can quickly and cost-effectively handle on each message, tool call, or sentence.
Jev is currently in the early trial phase and is available via API. TypeSafe AI positions it as a "System One" model—designed for fast, structured decision-making within software, rather than for human conversation generation.
Model URL: https://openrouter.ai/~typesafe/jev-latest
JevTypeSafeAIAIModelStructuredDecision-Making
This article is from AIbase Daily
Scan to view
Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications.
—— Created by the AIbase Daily Team
© Copyright AIbase Base 2024, Click to View Source -