Vision Models in Production: What Works

By C.W. Jameson · Published 28 March 2026 · Last reviewed 20 April 2026

Vision models handle PDFs and screenshots well. They fail predictably on handwriting and dense tables.

Practical lessons from deploying multimodal models: GPT-4V, Claude vision, and Gemini 1.5 Pro on real document workloads.