Dev Tests Jev Classifier: 50/50 Invoices Correct at 1/110th the Cost of Claude Opus

After projects built on Jev jumped from 46 to 160 in three days, a developer independently benchmarked the classifier on 50 multilingual invoices with OCR errors, achieving 50/50 accuracy at about $0.025 per decision—roughly 1/110 the cost of Claude Opus—though it confidently errs without explicit rules.

2026-09-19 ~ 2026-09-19 · 3 related posts

• Independent Jev test: 50/50 on hard invoice sorting at $0.025, but confidence misses rule errors — PawelHuryn · 2026-09-19

2 near-duplicate retellings: PawelHuryn · PawelHuryn