Opus 5 model card shows 5-agent coding teams reach 0.6 score 2.2× faster
OfirPress · x · 2026-07-25
A ProgramBench experiment from the Claude Opus 5 model card compares three ways to run multi-agent coding tasks: a single agent, a 5-agent team, and an async setup with background subagents.
Key results:
- The 5-agent team reaches the same 0.6 score about 2.2× faster than the single-agent baseline.
- The async setup is slower at the start because the lead agent must decompose the task first.
- It ends with the highest final score, while the 5-agent team gets to roughly 60% hidden tests passed the fastest.
The takeaway is a practical tradeoff: more agents can improve latency and/or final score, but the best structure depends on whether you optimize for early progress or end quality.
More from coding & agent
- Ben Bajarin: Frontier AI is becoming the enterprise cyber-defense stack — BenBajarin · 2026-10-26
- Building AI bots on raw APIs? Devs say models are 'dumb without the harness' — madeofAjala · 2026-10-26
- HarnessMM cuts agent tokens per task by up to 88% while raising success rates — omarsar0 · 2026-10-26
- The CLI is no longer the ideal interface for AI agents, argues AI communicator Elvis Omarsar — omarsar0 · 2026-10-26
- Microsoft demos intent-classifying agents built with Copilot App and Foundry, deployable across Teams and Outlook — marlene_zw · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26