The Sept. 15 release provides typed answers with probabilities instead of open-ended prose to optimize for deployment speed. In the JevBench composite score, which blends intelligence, calibration, and cost, Jev outperformed GPT-5.6 Luna despite the latter having higher accuracy in hard cases. LangChain analyzed the model to see if System One architectures provide a more repeatable approach to agent evaluation.
Wednesday, October 21, 2026
Story
Typesafe AI Jev Model Beats GPT 5.6 Luna in Composite Decision Benchmark
AI31 days ago
Typesafe AI's Jev decision model processed bounded software decisions more than 5 times faster than competing large language models. Tests by OpenRouter using Ori Eval showed Jev's slowest requests beat the median speed of other models, while its operational cost ranked as the second lowest behind Qwen3.8 Flash.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.