Jev vs black-box models
Most production black-box models hide weights and internals. Jev is the same class: hosted, not customer-LoRA’d, RLCD-trained per TypeSafe. Your control surface is the request schema, not a checkpoint.
Unofficial honesty page. Open replicas are not Jev — OpenJev. docs.typesafe.ai.
Comparison scope
If “black box” in your RFP means “we cannot audit weights,” Jev does not fix that. If it means “we cannot constrain outputs,” System One does constrain to your schema — still not factual proof.
Criteria that decide the architecture
| Axis | Jev (System One) | Black-box models |
|---|---|---|
| Weights | TypeSafe’s | Various |
| Output schema | Your questions | Often free text |
| Per-account train | No (vendor) | Sometimes |
| Open weights clone | Not Jev | Your risk |
Decision quality and control
Schema constraint reduces format surprises, not label errors. Treat it as a narrower box, not a glass box.
Integration trade-offs
Design model-agnostic actions (model-agnostic design) so you can swap Jev for a classical model later.
TypeSafe’s public models page lists jev-1.13 at $0.042 per million input tokens with output tokens free — a vendor claim, not a jev.pro measurement. Confirm on the models page before you bid.
When each approach fits
Prefer Jev when
- You want a hosted typed decision box
Prefer Black-box models when
- You must own weights or run offline
Black box + typed I/O + your traces. Open clones ≠ TypeSafe.
Schema you wrote, weights you do not own
Most production models hide weights. Jev is that class: hosted, not customer-LoRA’d, RLCD-trained per TypeSafe’s public pages. Your inspectable surface is the question pack, not a checkpoint. Sketch: version questions in git → pin jev-1.13.0 → log response.model, full probabilities, and the action → design model-agnostic so a classical model can replace the door later. Open replicas on this site are not Jev — OpenJev.
Paying the vendor-claimed $0.042 / M input (output free) does not purchase weights. Jaggedness still applies: a typed Noul can be confidently wrong. Schema-safe is not a glass box and is not factual correctness.
How “we can see the questions” gets oversold
- Format constraint as audit of weights. You constrained outputs. You did not open the network. If the RFP said “we must inspect parameters,” Jev does not satisfy that clause.
- Open-weight clone as drop-in. Third-party interface clones are not endorsed and are not TypeSafe Jev. Different calibration, different failure modes.
- Generated rationale as interpretability. Jev does not generate a why-essay. Do not bolt an LLM “explanation” onto the answer and call it XAI — see vs explainable models.
- Per-account training fantasy. Public docs we used: no customer LoRA. If you need a private head, that is another stack.
- One mega-Choice as the whole policy. Official primer: atomic questions. A 40-way Choice is still a black box with worse criteria.
When not to use Jev here
Do not use Jev if you must own weights, run fully offline, or satisfy a regulator who wants coefficients. Prefer a local classical model or a licensed checkpoint. Prefer Jev when you want a hosted typed decision box and can live with an inspectable request, not inspectable weights.
Write actions against the enum, not against “whatever Jev is this week.” That is the only leverage you have on a hosted box.
What this page does not claim
-
No architecture speculation beyond TypeSafe’s public pages.
-
Schema-safe output is not the same as factual correctness.
-
No independent bake-off numbers live here.
FAQ
Is Jev open source? The hosted model is not. Third-party interface clones exist and are not endorsed.
Can I download weights? Not from this site and not as TypeSafe Jev in public docs we used.
Is schema-safe the same as true? No. Out-of-schema is blocked. Wrong-but-in-schema still happens.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.
Hub: Comparisons. Siblings: vs explainable models, openjev, model agnostic design. Canonical: https://docs.typesafe.ai.