AT&T Cuts AI Costs With Model Routers and More Open-Source Models
AT&T reported that using LiteLLM model routers to match AI tasks to cheaper models reduced AI performance by about 2%, based on comments from an AT&T vice president overseeing employee AI use. The company plans to raise open-source or open-weight model usage from 40% to 60%–70%, using models from Nvidia, Meta and Google, while evaluating risks of DeepSeek and Moonshot.