Comparisons · Last updated 2026-09-21

Jev vs structured outputs

Structured outputs (JSON schema, tool-call arguments, grammar-constrained decoding) make an LLM emit tokens that parse. The model is still generating. You still pay output tokens. You still wait on a decode loop. You may still get a schema-valid object that is semantically wrong.

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.

Jev does not emit the document. You declare questions. You receive typed fields and distributions. TypeSafe’s launch post says they also wrap LLMs in a “System One LLM wrapper” to force structured decisions for comparisons — and that this wrapper is slower and more expensive than giving decisions without generating probabilities as text.

Criteria

When structured outputs remain the right tool

• You need a record with open string fields (names, emails, free comments).
• You need a reply the user will read.
• You need code or SQL.

Jev can choose among candidate strings you already extracted. It should not invent the strings. That is the jaggedness “Generation” section.

When Jev is the better shape

• Closed-set classification with a distribution.
• Rubric scoring with a legend.
• Yes/no with a probability.
• Many such judgments on one state.

You can use both: Jev routes or verifies; an LLM fills the open fields.

See also

• Comparisons hub
• vs JSON mode
• Jev vs LLM
• Extract as a pattern

When not to use Jev

Do not use Jev to fill a JSON Schema with open fields. Grammar-constrained decode exists for that. Jev will not invent an email address or a SQL string.

Do not use the TypeSafe “System One LLM wrapper” as if it were Jev. The launch post says that wrapper is slower and more expensive; it is for comparisons, not a substitute.

Do not treat a schema-valid tool-call blob as a Jev Choice. Tool calls still generate argument tokens. Choice returns a label plus a distribution.

Skip Jev when the only requirement is “parseable tokens” and you already own the generator’s retry loop.

Pairing-specific failure modes

• Valid-but-false records. Both stacks can emit a well-typed wrong answer. Structured outputs hide it behind parse success; Jev hides it behind a legal key. Review on confidence.
• Date/number slots. “expiry”: “next Friday” is a date library problem. Extract, then compare in code.
• Chained Choices as a writer. Spelling a payload one enum at a time is generation. Official jaggedness says not to.
• Wrapper mix-up. Teams benchmark the LLM wrapper, then quote those numbers as Jev. Label vendor claims; do not invent benches here.

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.

Production sketch

1. Split fields: closed enums / scores vs open strings.
2. Closed: Jev questions on filtered state. Pin jev-1.13.0. Code applies τ.
3. Open: LLM structured outputs (or JSON mode) for names, comments, SQL.
4. Optional: Jev citation-style check that a generated claim matches a span you kept.
5. Persist the record only after code says both sides cleared. Log response.model and the schema version of the generator.

FAQ

Do I pay Jev output tokens?

Vendor list price: output free, input billed. Confirm the models page.

Is the wrapper Jev?

No. TypeSafe says it is slower/more expensive. For comparisons, not a substitute.

Tool calls vs Choice?

Tool calls still generate argument tokens. Choice returns a label + distribution.

Schema-safe means true?

No. A grammar can force keys and still encode a false fact.

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.