SYLIN

Jev

by Typesafe · Model · Language & code models

promising

Cheap, fast decision model; rivals now match or beat it, mostly by their own claims.

TypeSafe's Jev is a System One model returning calibrated confidence over predefined options in 70-500 ms; waitlist API only

Why we say so

Almost everything here comes from reviewers repeating specs or from rivals' claims; nobody reports trying Jev hands-on. Simon Willison lists it as the cheapest decision model at 4.2 cents per million input tokens. Artificial Analysis measured it as faster (92 ms vs 211 ms per task) and cheaper than Google's Gemini Decision 1. Google says its model wins 5 of 7 shared benchmarks, and Microsoft and an HN commenter claim wins over Jev. Access is waitlist API only.

Evidence

  1. Nov 11
    Priced at 4.2 cents per million input tokens, the cheapest decision model; text and JSON input only
  2. Sep 27
    System One decision model from last week, used as the CLM comparison
  3. Sep 20
    Claimed frontier-level accuracy on structured decisions at far lower cost; '0% hallucination' only because answers are fixed options

Alternatives

Gemini DecisionOn average scores better than Jev and other models on decision benchmarks
LayaSelf-reported benchmarks slightly above Jev, faster and multilingual
Bespoke NimblePerforms pretty closely to Jev according to the video
CLMClaimed about 9x faster than Jev while matching its performance most of the time
Perplexity DeciderBeat Jev 1.14 and Decider v1.2 on an internal entity extraction/matching eval
MAI DecisionClaimed ahead of Jev, Google and OpenAI on the Artificial Analysis Decision Index
Built on Jev