Jev vs dJev
dJev is a hosted decision model built for speed and images. Jev is a hosted text decision model for choice, score and yes/no — with a calibrated probability on every answer. What differs, when each fits, and what it costs. Independent hosted gateway; not the official TypeSafe site.
What dJev is
dJev, from Maisa, is a hosted decision model based on DiffusionGemma. It accepts images as well as text, and its public materials emphasize short latency. It is a separate product from Jev, with its own account, its own pricing and its own preview terms — signing up for one does not sign you up for the other.
What you are choosing between
Jev stays on text. One request carries a shared state and up to eight named questions — choice, score or yes/no — and every answer comes back inside the option set you declared, with a probability. If the decision is about a picture or a camera frame, that is dJev's home ground; read its own notes on whether those probabilities are calibrated before you let it auto-approve anything on your side.
How the requests compare
Both speak a typed-decision shape, so the architectural move is the same: your state stays yours, your criteria are policy you author, and the model returns bounded answers. The operational differences are account and billing (dJev has its own), data path (images put camera and screen content in scope), and access. Optio gives you Jev against a Bearer key: 1 credit per 1,000 input tokens, credits that never expire, and a free playground to validate the wording on first.
What this site runs
Optio calls Jev. We do not resell dJev, and this page is not dJev's site. If the decision needs an image, call dJev directly. If your code needs a text decision it can threshold — routing, risk scores, policy checks — call Jev here.
Strengths and honest limits
Jev's limits are the ones worth knowing before you build: it reads text only, takes up to eight questions per request, and a state ceiling of roughly 100,000 characters sits in front of a 32k-token upstream context. Its strength, when the input is text and the decision must be thresholdable, is that every answer carries a probability over your own option set — and the failure path is an error shape with a request_id, not a plausible-sounding wrong string.
Quick comparison
| Question | Jev | dJev |
|---|---|---|
| Inputs | Text (string, array or JSON object, up to 100k chars) | Images and text |
| Answers | Choice / score / yes-no, probabilities over your declared options | Typed decisions; calibration claims vary by modality |
| Pricing | From $10 for 100,000 credits, never expire, 1 credit ≈ 1,000 input tokens | dJev's own account and pricing |
| Try it free | Playground: 30 runs/day anonymous, 50 signed in + 5 API credits on signup | Preview terms on dJev's site |
FAQ
- Is Jev or dJev cheaper?
- For text decisions on Optio: 1 credit covers a typical 400-token request, $0.0001 scale at Starter. dJev is priced separately and separately on its own site, so the honest answer is: price your workload both ways. The structural question is whether you need images — if not, per-decision text pricing is hard to beat.
- Can I use dJev and Jev in the same pipeline?
- Yes — they are different endpoints. A common pattern: dJev judges the screenshot or document image; Jev makes the routing, scoring and policy decisions over the text your system extracts. Keep each decision where its input type lives.
- Which one gives calibrated probabilities?
- Jev returns a probability over your declared options on every answer, and confidence alongside choice and score — that is the point of the typed answer space. For image decisions you should verify calibration on your own data with whichever model you pick; no model's calibration claim survives contact with every task untested.
- Is this the official Jev site?
- No. Optio is an independently operated gateway for calling Jev and is not affiliated with TypeSafe or with any alternative model mentioned on this page.