Threads and posts from Hacker News, Reddit and X. They are collected here because arguments surface the honest limitations faster than announcements do: scepticism about the speed and price claims, comparisons to BERT and to structured outputs, and people reporting what broke when they tried it. Nothing here is endorsed, only recorded.
23 discussions
Showing 23 of 23
The 1,863-point, 490-comment launch thread. Top comments argue the frontier-model framing is misleading, dispute the cannot-hallucinate claim on the grounds that type safety is not factual correctness, and note grammar-constrained decoding on ordinary LLMs already covers much of the interface.
A short video demo asking whether Jev understands colour, posted by generative artist Matt DesLauriers, who calls the model a leap forward for building new UX and UI paradigms on top of language models.
Elixir Forum thread (8 replies, 281 views) announcing an Elixir SDK for Jev with strict noul/choice/score constructors and evaluate/evaluate_many/streaming. Notes five independent Elixir Jev clients appeared in the same launch week.
A researcher claims prior open-source work with the same decision-only design and links the model, dataset, and paper. Companion to the Hacker News item 49736660.
Argues tool calling decomposes into classify plus action, and that pipelines switching between classification, structuring, deterministic code, and small agent loops make Jev a useful building block.
86-point, 12-comment thread on browser-use's agent. Commenters dispute the 7.1s Google Flights timing because timing starts after initial page observation, report demo failures, and object to paying per call for a cloud decision model.
36-point, 9-comment priority claim: the poster says he published a non-autoregressive, schema-constrained probability model in March 2025 (arXiv 2503.23303) using PPO over sequence embeddings rather than a parallel sampler. Commenters reply that the difference was marketing reach, not the idea.
Widely-shared skeptical-but-interested summary, around 145k views, of Jev's claims: 20-200x faster, 40-400x cheaper, $0.042 per 1M input tokens with free output, trained with RLCD over two years in stealth.
157-point, 23-comment thread on vinnylarouge/jevlike, an independent one-pass architecture that scores a changing list of text options against context. The submission claims only the same input/output shape as Jev, not equal quality or a reproduction of RLCD.
The founder's launch thread, around 63k likes, arguing that RLCD-trained decision models are a shorter path to economic value than chat models. Submitted separately to HN as item 49716682.
Company launch post announcing early access to Jev.
Public Convex-backed demo inviting a million judged questions instead of generated answers.
Vercel's eve agent engine announcing criteria-based model routing built on Jev; eve ships typesafe-ai/jev as the default evaluation model in its experimental evaluate path.
Author's announcement of the jevlike repo, which lets others train their own one-pass option-scoring models inferred from Jev's published input/output type.
Test report using Jev to check each agent action before it executes, reporting most attacks caught with few false blocks and much lower latency than an LLM guard.
Head-to-head claim that Jev beats Claude Opus 5 at move selection in Pokemon Showdown.
Builder reports that a small MLP head trained on top of Qwen 4B already reproduces Jev-like typed decision behaviour, used as an argument that the architecture is not a moat.
Demo in which Jev decides which model should serve a request before the request is forwarded.
Work-in-progress demo of Jev driving a Minecraft agent, including fleeing zombies at night, as a test of fast structured decisions in a game loop. URL taken from a fetched awesome-jev README; thread not opened.
Thread where builders describe using Jev as an agent tool-use safety layer and plan a prompt-complexity model router on top of it. URL taken from a fetched awesome-jev README; thread not opened.
Launch thread framing Jev as a low-hallucination, low-cost decision model for software rather than a chat model. URL taken verbatim from a fetched awesome-jev README; the thread itself could not be opened, so score, author and date are unconfirmed.
Argues that the purely engineering decisions inside security workflows belong to a typed decision model like Jev rather than a chat model.
Explains Jev's product shape as a classifier that takes its categories as input: context plus N options mapped to one probability each in a single pass, with no autoregression.
Nothing matches those filters
Clear a filter to widen the search, or send us the thing you were looking for.