TLDR Tech / IT
Test-driving Jev on a security task: identity resolution
Vincenzo Iozzo evaluated Jev, a specialized classification model, for identity resolution across five real organizations, achieving an F1 score of 0.948 in large organizations at a cost of $0.62 per 1,000 accounts. Jev outperformed Claude Haiku 4.5 (0.894 F1, 19x cost) and Sonnet 5 (0.911 F1, 47x cost), also demonstrating significantly lower latency. This suggests that smaller, calibrated models are more effective than prompted frontier LLMs for bounded, high-volume security classification tasks.
Read the full story