Metastone Launches TokenFleet Scheduling Framework

Compute operator Metastone unveiled TokenFleet at the 2026 CCF National High Performance Computing Conference. The framework targets cluster-level problems for LLM inference, offering predicted-latency routing that cut P95 time-to-first-token by 32.4%. On-target output throughput rose about 29.4%, while total output throughput increased about 1.5% in projected estimates.