Multi-agent coordination on TraceFix Coordination Benchmark 48 tasks
89.4Average Simulation Success RateTopology-monitored
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Topology-monitoredArchitecture=Topology-monitored, Backbone Model=gpt-5-mini2026.05 | 89.4 | 81.5 | 8.8 | 3.2 | |
| Prompt-onlyArchitecture=Prompt-only, Backbone Model=gpt-5-mini2026.05 | 86.2 | 69.7 | 21.1 | 7.3 | |
| Mediator-enforcedArchitecture=Mediator-enforced, Backbone Model=gpt-5-mini2026.05 | 78.4 | 44.7 | 4.5 | 1.9 | |
| Chat-onlyArchitecture=Chat-only, Backbone Model=gpt-5-mini2026.05 | 75 | 29.3 | 32.6 | 61.2 |