Split Vision and Reasoning Tasks to Cut Claude API Costs by 80%

A developer found that using Claude Haiku for OCR and Sonnet for extraction reduced per-document costs from ~$1.10 to ~$0.18, an 84% savings. The approach routes image-heavy OCR to the cheaper Haiku model while reserving stronger models for reasoning. This pattern can be applied to any pipeline where transcription and analysis are separate tasks.