Paul Couvert(@itsPaulAi)

This might be big Instead of just sending one prompt to multiple agents and comparing the results m...

8.5内容质量
This might be big Instead of just sending one prompt to multiple agents and comparing the results m...

TL;DR · AI 摘要

Offloop团队开发的D1层在多代理系统中实现SOTA性能，仅使用3B参数。

核心要点

  • D1层通过动态决策机制实现多代理协作，参数规模仅3B
  • 在GDPval基准测试中超越Claude code和Codex，覆盖2.4万亿美元就业场景
  • Offloop团队提出新型代理路由架构，参数效率提升显著

结构提纲

按章节快速跳转。

  1. §技术突破

    介绍D1层在多代理系统中的创新架构设计

  2. ·核心机制

    D1层通过动态决策机制选择最优代理执行

  3. ›参数效率

    仅使用3B参数实现SOTA性能，显著优于现有方案

  4. ·基准测试

    在GDPval基准测试中超越Claude code和Codex

  5. ›应用价值

    覆盖美国2.4万亿美元规模的就业场景

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）
  • D1多代理架构
    • 核心机制
      • 动态决策路由
      • 3B参数优化
    • 技术优势
      • SOTA性能
      • 成本效益
    • 应用场景
      • GDPval基准
      • 2.4万亿美元就业领域

金句 / Highlights

值得收藏与分享的关键句。

#多代理系统#SOTA#参数优化#Offloop#AI架构
打开原文

Paul Couvert on X: "This might be big Instead of just sending one prompt to multiple agents and comparing the results manually... this D1 layer is: - Deciding which agent acts at each step - Only 3B active parameters - SOTA on router benchmarks So it just acts between you and the different agents Pretty cool!" / X

Paul Couvert

@itsPaulAi

This might be big Instead of just sending one prompt to multiple agents and comparing the results manually... this D1 layer is: - Deciding which agent acts at each step - Only 3B active parameters - SOTA on router benchmarks So it just acts between you and the different agents Pretty cool!

00:00

Offloop

@Offloop

9h

Introducing Offloop! We're a team of four. Today our multi-agent harness hit state of the art on GDPval, ahead of Claude code and Codex across jobs that pay $2.4 trillion a year in the US. Offloop gives every knowledge worker what the Fortune 500 spends billions on: a

Show more

Paid partnership

5:09 PM · Jul 23, 2026

7K

Views

2

3

19

21