# Offloop 发布 D1 调度模型，解决多智能体通信混乱

- 来源：Chubby♨️ (@kimmonismus)
- 发布时间：2026-07-24 02:23
- AIHOT 分数：37
- AIHOT 链接：https://aihot.virxact.com/items/cmrxubmsp02neroxp9omqivpo
- 原文链接：https://x.com/kimmonismus/status/2080358121369739489

## AI 摘要

Offloop 推出 D1，一个轻量级调度模型，能在多智能体协作中决定每一步由哪个 Agent 执行、何时继续、何时停止或接入人工。在过夜基准测试中，D1 自动将两个卡住的任务在凌晨 2 点升级给人类处理。Offloop 的多智能体框架在 GDPval 上达到 SOTA，超越 Claude Code 和 Codex。

## 正文

Most multi-agent setups die the same way： every agent talks at once and the channel turns into noise.

Offloop built a dedicated layer for this. D1 is a small dispatcher model that decides at every step which agent moves next， when to keep going， when to stop， and when to pull a person in.
During their overnight benchmark run two tasks got stuck， and D1 escalated both to a human at 2am instead of grinding on them.

### 引用推文

> Offloop：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....