AutoTrust opens 27B decision model

AutoTrust AI released JEV-27B, a 27-billion-parameter Apache-2.0 model built for structured decisions inside self-hosted agent workflows. The release announcement says the model returns labels, scores, or yes/no decisions while retaining its reasoning path, and publishes weights at Hugging Face’s JEV-27B repository. Teams designing agent routing can file this as a typed-output alternative to paying a general model to generate prose; the archive’s JEV cost analysis covers the same decision-versus-generation boundary.