Cost-Optimal Foundation Model Deployment Portfolio for Transportation Management

arXiv cs.AI · 18 hours ago

Researchers formulated the Foundation Model Deployment Portfolio problem to determine optimal placement of language and vision models across transportation management tasks, treating it as a cost-minimization optimization problem subject to quality and latency constraints. A case study of five transportation functions found a mixed deployment strategy costing $34/month, 97% cheaper than using only closed-source APIs, by routing most functions to open-source models and one to closed APIs. The analysis shows on-premise GPU investment becomes cost-effective only above 309 vision queries per hour or if API pricing doubles.