CommunityJev Guides & Articles
Jev vs an LLM on support triage
OpenRouter times Jev 1.13, GPT Luna, and Claude Opus on 60 support tickets: intent plus an escalation flag, then says the reply still belongs to a generative model.
- Format
- Article
- Published by
- Community
- Author
- Kenny Rogers
- Last verified
- Sep 2026
- Platforms
- OpenRouter
Highlights
- 60 tickets, five intents plus escalation. Jev 1.13 got intent 59/60 and escalation 60/60.
- GPT Luna matched that intent score. Opus got intent 60/60 and escalation 59/60.
- Jev median 0.194 s and $0.001489 for the 60. Opus median 1.957 s and $0.17281.
- A second task uses 40 examples. Both sets are one day and one prompt.
Watch out
The post calls this the shape of the tradeoff, not a leaderboard. Jev cannot write the reply.
Questions about this entry
Short answers drawn from the linked project or page.
Barely, and the post admits it, framing the run as the shape of the tradeoff rather than a leaderboard: 60 and 40 examples, one day, one prompt. On the 60, Jev and GPT Luna both took 59/60 intents and 60/60 escalations while Opus went 60/60 and 59/60 — a one-ticket spread.
Per thousand triaged tickets: $0.0248 for Jev, $0.0921 for GPT Luna, $2.8802 for Opus. Jev sends more input tokens because every request carries the full criteria, and still lands at roughly a quarter of Luna and under one percent of Opus.
No. It takes text only, reads criteria literally, loses accuracy when the state carries irrelevant detail, and is unreliable at arithmetic, counting, and date comparison. The post's own pattern is to route and verify with Jev, then let a generative model write the reply.
From your own data, not from this run. Escalation comes back as a single yes-probability that you threshold yourself, and the post recommends picking that cutoff empirically. With 60 tickets there is no evidence here for any specific number.
Typed answers built from three primitives: choice picks one option from a set you define, noul returns a yes-probability, and score places content on levels you describe. The intent arrives as a choice with probabilities per option and escalation as a single number you threshold, so there is nothing to parse.
Jev takes text only: strings, JSON, or arrays of text, up to 64k tokens per request, with 32k for state plus the longest question. On the 60-ticket run its median latency was 194 ms and p95 633 ms. Images, audio, and PDFs are not accepted.
Send each ticket to the OpenRouter decisions endpoint, POST /api/alpha/decisions, with the model typesafe/jev-1.13. The LLM arms got the same definitions and escalation rule in a system prompt at temperature 0 and a request for a bare JSON object, one request per example.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
YouTube: Jev explained for routing and judging, with a comparison against LLMs.
Hacker News: Vercel on using Jev for defined decisions and GPT-6 Astra when the job also has to write or use tools.
flashbrew4 pointsTypeSafe AI Jev vs. GPT-6 AstraRead the thread on Hacker NewsHacker News: ASTGL replay chose local Laya routing over hosted Jev on one Mac Studio setup.
Jmeg8r15 pointsDoes an open-weight decision model beat a hosted one? Jev vs. LayaRead the thread on Hacker NewsMore Jev resources like this
Vercel's guide for when a typed Jev decision is enough and when a job still needs GPT-6 Astra to write, see images, or call tools.
Jev Guides & ArticlesArticle#community#article#routing
Community
Vercel's guide for seven jobs Jev can take: routing forms, prioritizing tickets, reviewing tool calls, picking a model, categorizing documents, moderation, and checking an answer.
Jev Guides & ArticlesArticle#community#article#routing
Community
Kenny Rogers explains Jev as a decision model and shows Choice, Score, and Noul responses from calls he placed through OpenRouter on 21 September 2026.
Jev Guides & ArticlesArticle#community#article#explainer
Community
Vercel's changelog for typesafe-ai/jev on AI Gateway, with AI SDK 7 experimental_evaluate and a support-ticket routing example.
Jev Guides & ArticlesArticle#community#article#vercel
Community
OpenRouter's Banking77 run: 3,080 support utterances, 77 intents, Jev 1.13 against Claude Opus 5 on the same one-line criteria, with latency, cost, and a confidence cascade.
Jev Guides & ArticlesArticle#community#article#evaluation
Community
Philipp Acsany's tutorial replaces a Y-or-N input check with Jev, then asks Noul, Score, and Choice in one call through OpenRouter.
Jev Guides & ArticlesArticle#community#article#tutorial
Community
From the Jev guides
Original write-ups on what this entry is for.
Confidence is not accuracy, and a single global threshold is rarely the right policy. A practical guide to per-action gates, measured thresholds, and the logs you will want later.
Jev Practices & Patterns#confidence#routing#evaluation
Read article
The typed-answer shape removes a category of data risk, not all of it. Zero data retention per surface, the legal set to read, and the controls that remain yours to run.
Jev Guides & Articles#privacy#compliance#legal
Read article
TypeSafe serves Jev from one API. A second tier of independent sites wraps it with their own keys, credits, and playgrounds. How to read the disclaimers before you paste a key.
Jev Guides & Articles#ecosystem#trust#hosting
Read article
From the Jev community
Posts from builders shipping with Jev right now.