Jev · 21 pages · indexed 2026-09-27
Every Jev guide, in one place
21 pages, grouped by the question each one answers: what the model is and what a decision costs, how people wire it into something that ships, what they use it for, and where the raw records live. Nothing here is new — every link below goes to a page that already exists, and this page exists so you do not have to open six of them to find out which one you wanted.
Model facts read from the public listings, 2026-09-18 Counts read from the same files the pages render New here? What Jev is, then how it differs from an LLM.
What Jev is, what it returns and what it costs
Start here if you have not made a call yet. The order is the order the questions come up in: what it is, what comes back, what that costs, and what we measured ourselves.
-
Jev, the model on OpenRouter
The front door: what it is, the listed price, and the launch post that announced it.
-
Jev, specified
The long version — checkpoints, question types, the API shape, and the thirteen adversarial cases we sent it.
-
Jev vs an LLM
The difference is not that one model is smaller. One produces words, the other produces a typed decision — and that changes where it goes in your program.
-
What a decision costs
Worked out two ways: the vendor's per-token list price, and what we actually paid for a 420-token call.
-
The catalogue report
What we counted and what we measured, with the JSON — the numbers on this page's own cards, in one place.
How to build with it
The wiring: what other people did when they put Jev in a loop, and what they gave up to do it.
-
How people actually use it
Eleven repositories read line by line — the action spaces, the fallbacks to text, and the sentence in each README that says what it does not do.
-
Jev with Claude Code
What the integration actually changes: who proposes, who decides, and which of the two you should let own the loop.
-
Open source, and what it gives up
The replicas, the ports and the wrappers — what each one reproduces, and the thing it cannot reproduce.
-
Docs, collections and repositories
Everything worth bookmarking, including the two collection pages that already curate this space.
29 records
What people use it for
One page per kind of decision, cut across every column — the same record can appear in two of them, which is what an axis is for.
-
Every use case in one list
Filter the whole corpus by what the thing decides, rather than by where it was published.
-
The jobs this creates
Roles people are actually advertising and hiring for around decision models, read off the postings.
-
Browser agents
Driving a browser: clicking, reading, deciding the next step.
21 records
-
Triage and routing
Picking a route, a label or a next model — classification where a wrong answer is recoverable.
40 records
-
Coding and context
Inside a coding agent: compaction, context windows, the harness around the model.
22 records
-
Research and data
Benchmarks people ran, architectures people reverse-engineered, results that did not work.
61 records
-
Media and content
Generating, sorting or reviewing words, images and posts.
24 records
-
Trading and markets
Decisions with money attached, on a clock.
7 records
Where the records are
The four columns plus the two we keep for context. Every row was read by a person before it was published; the counts are what is live on them today.
-
What people built
Repositories with a dossier — what it does, how it works, what we checked, and what nobody checked.
22 records
-
What X is saying
Posts read in full, each with a sentence of ours and the author's own numbers where they gave them.
101 records
-
What people filmed
Walkthroughs and demos, with view counts and durations read from the platform.
15 records
-
What Hacker News made of it
The launch threads and the independent write-ups, including the sceptical ones.
45 records
What this page is not
It is not the index of everything: the four columns carry 5 shelves of records, and the use-case facets are a second cut across them. Those are in the directory below, with their live counts. This page is only the reading — the pages we wrote because a question needed an answer longer than a caption.