TypeSafe's Jev decision model costs $0.042 per million tokens, matches Sonnet 5 accuracy

In TypeSafe’s internal workflow benchmarks across security, agent observability, invoice processing and customer service, Jev achieved 67.8 % mean accuracy at $0.0004 per case and 0.4 s latency, matching Claude Sonnet 5’s accuracy while costing far less. The top frontier model (“sol”) still led with 74.1 % accuracy but was slower and pricier. Early adopters such as Vercel AI Gateway saw nearly 13 % of paid teams use Jev within 24 hours, double the share of GPT‑5.6 and six times that of Fable 5.1.

Fastino Labs quickly followed with GLiDE (released Sep 30) and GLiNER2.5‑Decide (Sep 24), while open‑weight reproductions like kev, OpenJev and JevK5 allow local deployment. Decision models are positioned for high‑volume, bounded judgments—routing, classification, triage, guardrails and evaluation—while traditional LLMs remain suited for open‑ended generation.