METASTONE launches TokenFleet, an LLM inference scheduler for GPU clusters
Overview
METASTONE, an independent compute operator, launched TokenFleet, a distributed scheduling framework that routes large language model inference requests across GPU clusters and data centers.
The company announced it at the 2026 CCF National High Performance Computing Conference.
METASTONE says its TokenPilot module cut P95 time to first token by 32.4% in tests on a GLM5.3 workload. It says its SLOMax module, in a revenue estimate, raises goodput about 29.4% by reducing timeouts, cancellations and duplicate computation. The company says none of these figures have been independently verified.
Written by AI from the articles below · updated Oct 21, 12:18 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- PandailyNewsMETASTONE launches TokenFleet to schedule LLM inference across GPU clusters and data centers
AIMETASTONE has launched TokenFleet, a distributed scheduling framework that routes LLM inference requests across GPU clusters and data centers to raise effective token output. In slides from the 2026 CCF National High Performance Computing Conference, predicted-latency routing on a tuned GLM5.3 deployment cut P95 time-to-first-token by 32.4%, though the company says none of its figures have been independently verified.
- QbitAINewsMETASTONE launches TokenFleet, a distributed scheduling framework for inference clusters
AIMETASTONE launched TokenFleet, a distributed scheduling framework for GPU inference clusters, at the 2026 CCF National High Performance Computing Conference. Its TokenPilot module reports a 32.4% cut in P95 time to first token in tests on a GLM5.3 workload. Its SLOMax module is shown in a revenue estimate to raise goodput about 29.4% by reducing timeouts, cancellations, and duplicate computation.
Heat Trend
Not enough continuous observations to show a trend yet.