Databricks adds Jev 1.14.0 to Unity Gateway as an externally hosted preview

Last updated:

Source · Model card: Jev

publicado (OK do Fabricio 18/10 BRT).

Databricks’ October 2026 release notes have three entries about decision models.

16 October: TypeSafe AI Jev 1.14.0 on Unity Gateway. The note says TypeSafe hosts the model, that an account admin must turn on the “Externally-hosted TypeSafe AI Models” preview, and that TypeSafe’s terms govern its use. So this is a route to TypeSafe’s own endpoint through Databricks’ gateway, not Jev running on Databricks hardware. The note gives no price and no latency; we publish none, and any speed or cost figure for this route should be read as an UnverifiedClaim until Databricks or TypeSafe documents it.

2 October: OpenJev (Qwen3.5 4B) as a pay-per-token service. Databricks says to use the TypeSafe System One API to evaluate yes/no, choice and scoring questions and get structured answers with probabilities. 9 October: the same model can be the evaluator in a custom LLM-as-a-judge service policy (Beta); Databricks says it “classifies content without generating output tokens”, so it typically responds faster than a chat evaluator. We have not matched “OpenJev” to a model in our catalog; Databricks does not link a repository.

Why it matters. It is the first big data platform to put a decision model inside its gateway policy layer, next to the Vercel AI Gateway and LiteLLM routes. Enterprise teams that already run Databricks can reach Jev without a separate contract path, though the preview flag and the TypeSafe terms still apply. Added to the distribution list on the Jev page; no score changes.