Jev vs JSON mode
JSON mode (and cousins: “return JSON only,” YAML fences, markdown tables) is a prompt-plus-parser contract. Failure modes are familiar: truncated objects, extra keys, strings where you wanted enums, retries.
This page is for builders crossing from curious to production-minded Jev use. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Cover intents; do not clone rival IA. Confirm facts on docs.typesafe.ai.
The TypeSafe HTTP response is JSON because HTTP APIs use JSON. The model is not being asked to author that document as a completion. Your client unmarshals a stable schema: model, answers, usage.
If you find yourself writing json.loads(completion) after a Jev call, you are on the wrong endpoint.
Independent projects that read option logits instead of generating JSON are exploring a related idea on open weights. They are not Jev. See OpenJev.
Criteria
| Axis | JSON mode (generative) | Jev (System One) |
|---|---|---|
| What the model does | Decode tokens that look like JSON | Answer typed questions |
| Parse risk | Truncation, extra keys, retries | Client unmarshals a stable HTTP schema |
| Open string fields | Possible | You supply candidates; Jev does not invent them |
| Output tokens | You pay the decode loop | Vendor list: output free, input billed |
| Schema-safe ≠ correct | Yes | Yes |
See also
When not to use Jev
Do not use Jev to emit a record with open strings (names, emails, free comments). That is JSON mode or structured outputs on a generator.
Do not use Jev to write the email, the SQL, or the JSON document the user will store. Jaggedness: generation. Do not chain Choices to spell a payload.
Do not treat “the HTTP body is JSON” as “Jev is JSON mode done right.” Different product: typed questions, distributions, no prose generation.
Skip Jev when you only need a parseable blob and you already tolerate retries and schema-valid nonsense.
When JSON mode wins
JSON mode wins when the artifact is the point: a config object, a tool-call argument blob, a row the user will edit. Jev wins when the decision is the point: which closed key, which score level, what probability.
You can compose: Jev chooses a template or a route; JSON mode fills open fields on an LLM.
Pin jev-1.13.0 after you fit floors. TypeSafe’s published list price for jev-1.13 is $0.042 per million input tokens (vendor claim — confirm on the models page); output tokens are free on that same page.
Pairing-specific failure modes
- Wrong endpoint habit. Teams wrap Jev in a “just parse the JSON” helper and then “fix” missing keys. There are no author-keys to fix.
- Date/number fields in the prompt. Asking either stack to compare amounts is jaggedness. Extract in code; decide on language.
- Schema-safe wrong enum. Jev cannot invent a key outside criteria. It can still pick a wrong key. Same honesty as JSON mode’s valid-but-false objects.
- OpenJev confusion. Reading option logits on open weights is not TypeSafe Jev and is not endorsed here.
Production sketch
- Split the job: closed decisions vs open records.
- Closed: POST typed questions to Jev. Pin
jev-1.13.0. Unmarshalanswers— do notjson.loadsa completion. - Code applies τ and
confidencefloors. Wrong or low → review, not a guessed key. - Open records: LLM + JSON mode / structured outputs. Optional Jev check on a claim vs a span before you persist.
- Log
response.modeland the criteria SHA. jev.pro does not sell, issue, or proxy API keys.
FAQ
Is Jev “JSON mode done right”?
No. It is a different product: typed questions, distributions, no prose generation.
Can I still use JSON mode for emails?
Yes — that is generation. Compose with Jev on the decision.
Schema hallucination?
Jev cannot invent a key outside criteria. It can still pick a wrong key. Schema-safe ≠ correct.
Does Jev charge for the JSON response body?
Vendor list price bills input tokens; output is free on the models page. That is not “JSON mode without output tokens” as a slogan — confirm current pricing there.
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.