AgentFleet cross-device scheduling benchmarked: Throughput and reliability performance

On November 5, 2026, at 22:53, an article titled「AgentFleet Benchmarked: Throughput and Reliability Across Devices」appeared on Juejin, presenting hands-on measurements of AgentFleet's cross-device AI fleet scheduling. The author had previously written about AgentFleet's architecture, in which each device is enrolled into a fleet that coordinates across devices; this piece, the author says, is about letting the data speak and testing whether「cross-device scheduling」is a real capability or just a slogan.

The article claims three reproducible numbers. On scheduling throughput, the abstract states 303 tasks/s, while the visible fragment points to a range of 303~377 tasks/s; the two versions of the figure do not match exactly and are reported here as published. On reliability, the abstract reports zero task loss when a device drops off the fleet during scheduling, a result the author presents as evidence that AgentFleet holds up in distributed, collaborative setups rather than only in theory. The excerpt supplied cuts off at the throughput item, so the full set of measurements behind those claims is not available here. At present, the review stands at this one piece of coverage: a single set of benchmark results published on a Chinese developer community site, with both the throughput and the drop-resilience figure coming from that one article, and the larger 303~377 tasks/s range only partially visible in the original fragment.