DeLM: decentralized multi-agent coordination gains up to 17.5 pts accuracy, 2.49x faster than Codex and Claude Code
StanfordAILab · x · 2026-10-09
Stanford AI Lab amplified an updated paper from Mao Yuzhen's team introducing DeLM, a decentralized approach to multi-agent systems.
- Core idea: drop the main agent in favor of a shared context and task queue; agents claim tasks asynchronously, build on or correct each other's progress, and see each other's status.
- Results: on long-horizon tasks from Terminal-Bench 4.0, DeepSWE v1.1, and ProgramBench, accuracy improves by up to 17.5 percentage points and runs up to 2.49x faster than Codex and Claude Code harnesses.
- Deployment: an open-source MIT-licensed plugin (delm-agent-plugin) lets you run collaborating DeLM agents directly inside Codex and Claude Code.
- Compared with the first version in early June, the updated paper expands evaluation beyond SWE-bench to the three new benchmarks.
Related event: Stanford's DeLM Boosts Multi-Agent Coding Speed 2.49x(4 posts)→
More from coding & agent
- Stop wasting Opus: a 5-layer Claude Code setup with Sonnet, Haiku and Opus on call — Arindam_1729 · 2026-10-09
- $200 Max plan greentext: agents burn 29% quota, touch zero files, run no tests — BLUECOW009 · 2026-10-09
- Claude Code 2.1.295 ships 143 CLI changes, adds blocking hooks and stream timeout controls — ClaudeCodeLog · 2026-10-09
- Claude Code 2.1.295 ships 143 CLI changes: failing hooks now block actions, gateway TTFB timeout added — ClaudeCodeLog · 2026-10-09
- Senzing ships Claude Code plugin bringing entity resolution to coding agents — JeremyCMorgan · 2026-10-09
- Dev adds railway 'pointing and calling' safety ritual to AI agent workflows — menhguin · 2026-10-09