Stop defaulting to your biggest model for everything. Meta's Muse Ember: $0.10/M input, $0.50/M output, 1M context. Meta reports 37.8% on AutomationBench-AA vs 13.9% for Claude Haiku 5.5. The playbook: route routine steps (classify, extract, summarize) to a cheap model, save the frontier model for hard reasoning. Many teams could cut inference bills this way. More at www.contentbuffer.com #AI #Meta #LLM #AIAgents #DevTools What share of your AI calls really need a frontier model?