Jev vs djev: Text Decisions, or Decisions About Pictures
This comparison has a shortcut. djev accepts images — including images as the options in a choice question, and frames sampled from a live camera. Jev reads text. If your decision is about a photograph, you are not choosing between them; you are using djev.
For everything else, the interesting differences are about maturity and operating posture rather than capability.
Background
What djev Is
djev is a hosted decision service with the same core idea: send context, declare yes/no, choice or score questions, get probabilities back. Its own site describes image input alongside text and JSON, images usable as choice options, and live camera sampling, with model processing on Modal and delivery on Vercel.
It publishes an input-token price with free output, in the same ballpark as the model behind this site. Its own documentation is careful about what the probabilities mean, noting they are model estimates rather than guarantees of correctness — which is the right thing to say and worth repeating about every model on this page, including ours.
What it does not publish, as of the date we checked, is the sort of calibration and latency detail you would want before putting it on a hot path at volume. That is not a criticism of a young project; it is a reason to measure it yourself rather than assume.
Source for the factual claims above: djev.dev, checked 2026-09-25. We link it so you can check it rather than take our word for it.
Side by side
What Actually Differs
Based on each project’s own published material, checked on 2026-09-25.
| Jev (via jev-ai.org) | djev | |
|---|---|---|
| Input types | Text and JSON | Text, JSON and images, plus live camera frames |
| Images as choice options | No | Yes |
| Question types | Yes/no, choice, score | Yes/no, choice, scored rubric |
| Billing unit | Input tokens; output free | Input tokens; output free |
| Hosting | Vendor-hosted model, resold through this site | Hosted service on third-party infrastructure |
| Published calibration detail | Vendor publishes calibration claims; we publish measured costs and latency | States probabilities are estimates, not guarantees |
How to choose
Neither One Wins Every Time
Choose the text model when
- Your state is text: tickets, documents, transcripts, rows, JSON payloads, scraped pages.
- You need the decision on a production hot path and want published limits, a documented error contract and a version you can pin.
- You want an account, keys with spend caps, request history and a batch workbench rather than only an endpoint.
- You would rather send a caption or an OCR transcript than the image itself, which is often the cheaper answer anyway.
Choose djev when
- The thing being judged is an image: a photo of damage, a receipt, a shelf, a document scan, a video frame.
- You want images to be the options, not just the input — "which of these three does this look like" is a shape the text model has no answer for.
- You are sampling a live camera and need a decision per frame.
- You are prototyping and its hosted preview is the shortest path to finding out whether the idea works at all.
We sell one of these two, so read the right-hand column with that in mind — and then run both on two hundred of your own labelled examples, which settles it better than any page on the internet can.
Questions
Jev vs djev: Common Questions
Can I make Jev work on images by describing them first?
Often, and it is worth trying. A vision model produces a caption or an OCR transcript once, and every subsequent decision is a cheap text call over it. That is a good architecture when many decisions share one image, and a poor one when the decision depends on detail a caption would drop.
Are the two APIs compatible?
They are conceptually similar and not wire-compatible. Expect to rewrite the request builder and the answer reader, not the surrounding design.
Which is cheaper?
The published per-million input rates are close enough that the difference will be swamped by how many tokens your payloads actually contain. Measure both on your real inputs; on this site every use-case page shows the measured token count of its example so you have something concrete to compare against.
More
Other Comparisons
Jev vs Laya
The open-weights alternative you run yourself, against the hosted model you call.
Jev vs SemIf / OpenJev
The self-hosted reimplementation that reads option logits straight from an open model.
Jev vs LLM structured output
JSON mode guarantees the output parses. It does not tell you how sure the model was.
The Cheapest Way to Settle It
Load a scenario, paste in your own text, and see what the distribution says. Editing costs nothing and needs no account.