TrueFoundry Auto Routing was 69% cheaper in an academic-prompt test, a user reports

A user working with TrueFoundry's Auto Routing argues that most AI requests don't need the strongest model. They say the tool classifies requests as simple, medium or complex and sends them to an assigned model through a single endpoint. Classification can use either a free heuristic classifier or an LLM classifier for more nuanced judgments. The post cites a TrueFoundry experiment comparing Claude Opus alone with routing across Haiku, Sonnet and Opus: on 550 deterministically graded academic prompts, it says routing was 69% cheaper while retaining 98% of baseline quality.