Start with the thing it contrasts against. An autoregressive language model produces text by predicting one token, feeding that token back in, then predicting the next. Each step waits for the one before it, which is why a long answer takes longer to return than a short one.
TypeSafe’s launch post describes Jev’s sampling as “Parallel”, which “Generates all outputs in a single query” versus sequential token generation, and calls it “Incredibly efficient and hardware-aware”. That is the whole public claim. Parameter count, base architecture, training data and the RLCD method are all unpublished, and the weights are not released, so nobody outside TypeSafe can check how the parallelism actually works.
Sean Goedecke makes the mechanical point independently: “If you want fast, parallelized structured output against limited choices, you don’t strictly need to do autoregressive generation at all… Since LLMs ingest all input tokens in parallel, this is way faster than generating the entire structured output.” Community guesses run to an encoder-style classifier head and to masked diffusion. They are guesses.
The number attached to the claim: TypeSafe reports 70ms to 500ms end to end for Jev against “3 to 329 seconds” for frontier models. No independent benchmark existed as of 2026-09-18. See what is public about Jev’s architecture and the typed output entry.