Nexos routes 84% of coding work to cheaper models

Nexos.ai says its smart router sent 84% of coding-agent requests to cheaper open-weight models and cut one production workload’s AI spend 59.2%, according to the company’s launch results. The router kept 16% of requests on frontier models for planning, while a second test reported a 60.4% reduction; these are company-run production tests, not independent benchmarks. Teams testing Copilot’s own model-routing tiers should compare accepted work and failure rates before treating the savings claim as a budget line.