Jev Leads a Push Toward Fast, Measurable Decision Models
36Kr reports that TypeSafeAI launched Jev in mid-September, a model designed to classify inputs into preset options and return boolean values, scores, or probabilities rather than conversational text. The report says Jev attracted more than one million users within weeks and helped popularize “decision models,” followed by OpenAI’s Decisions API and similar products from AWS, Microsoft, and Cloudflare; these adoption, valuation, and competitor claims come from a single source and remain unverified. It also argues that AI evaluation is expanding beyond generated-content quality toward decision accuracy, latency, and execution cost.
The report portrays decision models as a specialized layer for frequent, constrained classification tasks in agent workflows, where directly usable outputs and low latency may matter more than fluent generation.
9月中旬,一家名叫TypeSafeAI的初创公司发布了一个叫Jev的模型。它不写文章,不聊天,甚至不输出一段完整的句子。它做的事只有一件:把输入归入预设的选项,返回一个是/否、一个评分、一个概率。就是这样一款“不会说话”的模型,上线数周用户破百万,财富500强中约三分之一的企业开始使用,24天内估值从约2亿美元飙到75亿美元。
它带出了一个迅速被行业共识的新品类名字:决策模型。OpenAI两周内跟进Decisions API,AWS、微软、Cloudflare相继入场,不到一个月涌现出十余款同类产品。决策模型的快速兴起,反映出人工智能产业正在发生一项重要变化:模型的价值评价,正在从内容生成质量进一步延伸至决策的准确性、响应速度与执行成本。
OmAI联汇CEO兼首席科学家赵天成指出,很多场景不需要夸夸其谈地生成内容,而需要一连串连续、快速、可度量的决策。“大语言模型其实也能做很多决策,但决策模型是快速模型,把全部能力和精力聚焦在可度量、可验证的决策上。决策模型切入的,正是另一类场景:这里需要的不是一段生成出来的内容,而是一个直接可用的判断——快、低成本、端侧可读。”
#TypeSafeAI#Jev#OpenAI#Decisions API#AWS#Microsoft#Cloudflare#OmAI联汇#VLX-Flow#VLX-Seek#VLX-VR#VLX-Go