독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
주제DisconfirmAI 편집부업데이트됨 2026-10-24 11:09

决策模型品类

6 단계4 건5 출처

【背景】继 TypeSafe 于 9 月发布仅输出判断而非文本的 Jev 模型后,“决策模型”迅速成为 AI 基础设施的新品类。OpenAI、亚马逊 AWS、蚂蚁数科等厂商在两周内相继入场,将高频分类、路由和评分任务从通用大模型中剥离。

这一趋势的核心在于成本与效率:Jev 定价低至每百万输入 token 0.042 美元,且输出免费;蚂蚁数科自报其 LingDT-Decision-1.0 模型 Token 消耗降低超 90%。然而,该品类的护城河尚存争议。一方面,Vercel 数据显示 Jev 上线 24 小时内即被 13% 的付费团队采用,证明需求强劲;另一方面,由于架构相对简单(如基于 Qwen 底座微调),OpenAI 仅用两周便推出竞品 Decisions API,引发关于“白菜价”竞争和快速迭代的担忧。中国团队如上海人工智能实验室和 StartLux 则通过开源和本地部署策略,填补了数据敏感场景下的空白,与云端闭源方案形成差异化竞争。

현재까지의 과정

  1. 2026-09-15

    前 OpenAI 研究员 Diogo Almeida 创办的 TypeSafe AI 发布 Jev 模型,主打低成本、高并发的结构化判断,不生成文本。

    geekpark.net ↗
  2. 2026-09-28

    AWS 工程师 Marc Brooker 博客披露个人项目 Hobson,验证小参数模型在决策任务上的可行性,随后演变为 Amazon Strands Decider。

    brooker.co.za ↗
  3. 2026-10-01

    亚马逊 AWS 正式开源 Strands Decider 2B,基于 Qwen3.5-2B 底座,支持本地部署,直接对标 Jev。

    techcrunch.com ↗
  4. 2026-10-02

    Databricks 宣布 ai_decide 功能,Perplexity 发布开放权重决策模型,行业跟进速度加快。

    cloudedjudgement.substack.com ↗
  5. 2026-10-22

    蚂蚁数科发布 LingDT-Decision-1.0(4B/27B),自报在多项基准中优于 Jev 1.13,Token 消耗降低超 90%。

    leiphone.com ↗
로그인 / 회원가입
독서리서치레이더투자 프레임워크
  • 2026-10-23

    OpenAI 向所有客户开放 Decisions API,利用 ChatGPT Luna 优化将推理延迟降至百毫秒级,加剧市场竞争。

    techcrunch.com ↗
  • 아직 해결되지 않은 문제

    • 决策模型的准确率是否真能显著超越通用大模型(如 GPT-6-Luna)在简单分类任务上的表现?目前缺乏独立的第三方对比数据。
    • 随着 OpenAI 等巨头入场,低壁垒的决策模型是否会迅速陷入价格战,导致初创公司(如 TypeSafe, StartLux)难以维持商业价值?
    • 本地部署模型(如 StartLux, Intern-Decision)在隐私敏感场景下的实际采用率能否抵消云端 API 的成本优势?

    출처

    • geekpark.netGeekPark: Jev 之后,中国团队开始深挖 AI 的「直觉层」↗
    • techcrunch.comTechCrunch: Amazon releases its own Jev clone↗
    • cloudedjudgement.substack.comClouded Judgement: Decision Models: The Next Shoe to Drop↗
    • leiphone.comLeiphone: 蚂蚁数科发布LingDT-Decision-1.0↗
    • techcrunch.comTechCrunch: OpenAI makes its Jev competitor widely available↗

    이 주제 커버리지

    • OpenAI Makes Decisions API Widely Available, Cutting Latency to Sub-Second Levels2026-10-23뉴스

      OpenAI has opened its Decisions API to all customers, optimizing ChatGPT Luna to reduce inference latency from seconds to milliseconds, challenging TypeSafe's Jev models.

    • Ant Digital Technologies Releases LingDT-Decision-1.0 Model, Claims Over 90% Token Cost Reduction2026-10-23뉴스

      Ant Digital Technologies launched the LingDT-Decision-1.0 model for rapid decision-making, claiming superior performance and significant cost reductions in enterprise AI scenarios.

    • GeekPark: From Jev to StartLux, How AI Decision Models Fill the 'Intuition Layer' for Local Deployment2026-10-22뉴스

      A GeekPark article analyzes AI decision models ("System One") represented by TypeSafe's Jev, which decouple high-frequency judgment from deep reasoning. It highlights that while the category has low barriers and rapid iteration, Chinese teams like Shanghai AI Lab and StartLux are filling infrastructure gaps for data-sensitive scenarios through open-source and localization strategies, enabling complex Agents to run on personal computers.

    • Decision models emerge as a category, with three major vendors launching this week2026-10-02뉴스

      Decision models that output judgments instead of text drew launches from OpenAI, Databricks and Perplexity in one week; pricing is far below general models, accuracy unverified.