MAC Bench - Eight Agents without Coordination: leaderboard
Multi Agent Coordination Bench with eight agents of one model and no way to exchange messages: the silent team in tasks whose environment splits what agents see or control (and in Dubai ATC), and the best of eight independent single agents in the other shared tasks and Power Outage (the best of four for Opus 5.5 in Cave Rescue); the gap to the team-of-eight board is what messaging adds or costs.
Metric: Mean final score x100, eight agents without messaging (silent team or best of eight single agents). Source: benchmarks.bespokelabs.ai. Saturation forecast: Around October 2027. 3 models tracked. Updated 19 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol | 47.86 | #15 |
| 2 | Claude Opus 5.5 | 47.61 | #9 |
| 3 | Gemini 4 Argon | 35.63 | #18 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=mac-bench-eight-agents-without-coordination · How It Works · Data refreshed daily, snapshot 2026-11-20.