Overview

Jev is TypeSafe’s System One decision model (e.g. jev-1.13.0) outputting typed Noul/Choice/Score answers with calibrated probabilities — no text generation, structurally no hallucination on predefined outputs.

Benchmarks (early adopter reports)

  • Vercel: 5–18× faster than LLM classifiers vs OpenAI Luna
  • Bryo AI: 10–20× cheaper than Gemini for routing tasks
  • Cloudflare AI catalog lists 32K context window

Benchmarks are early adopter/vendor reports, not peer-reviewed. Base architecture suspected to build on open-weight LLM backbone.

Sources