Jev vs Structured Outputs

What Structured Outputs is

A feature of chat models (GPT-4o, Claude Sonnet, etc.) that guarantees the response conforms to a provided JSON schema. The model generates arbitrary JSON matching your schema from unstructured input.

• Output shape: arbitrary nested JSON, defined by your schema
• Schema guarantee: response conforms to your provided JSON schema
• Use case: structured extraction — parse receipts, label screenshots, extract fields
• Latency: chat model generation time, typically seconds for complex schemas
• Pricing: chat model rates, $0.30–3.00/M input depending on model
• Probabilities: optionally returns token-level logprobs; no per-option probabilities for typed decisions

What Jev is

TypeSafe's Jev is a decision model (System One), not a chat model that extracts or generates. It answers typed questions with finite pre-defined options.

• Question types: Choice (pick one), Score (where on a scale), Noul (probability 0–1)
• Returns: per-option probabilities and confidence score
• Input: text only (string, JSON object, or array of strings)
• Use case: high-frequency classify/route/act — agent next action, model routing, inbox triage, scoring feeds
• Latency: 70–500 ms per TypeSafe docs
• Pricing: $0.042 per million input tokens, output free
• Batch: ~60 questions cost about the same as one

Source: TypeSafe models

Side by side

Not the same category. Structured Outputs is a chat model feature; Jev is a purpose-built decision model. The table clarifies where each fits.

When to use Structured Outputs

Complex schema extraction. Parse a receipt with nested line items, label a screenshot with complex annotations, extract fields from a document where the schema has multiple levels. The output shape is complex or nested, not a simple typed question.

Occasional workflow. The extraction runs a few times per user or per hour, not thousands of times per minute. Chat model pricing ($0.30–3.00/M input) is acceptable because the value is in the extraction quality, not in per-call cost.

Generation required. The model must generate structure, not pick from pre-defined options. You cannot list all possible outputs ahead of time.

When to use Jev

High-frequency classify/route/act. Agent next action (which tool to call), model routing (which tier/model), inbox triage (which queue), scoring feeds (is this post viral), tool selection. The question is which option from a finite list you provide.

Probabilities and confidence required. Your code acts on confident answers automatically and escalates uncertain ones to a person or larger model. Structured Outputs does not return per-option probabilities for typed decisions.

Multi-question batching. You need to ask many questions in one call (e.g., score every item in a feed against 8 criteria). Jev batches ~60 questions at the cost of one; Structured Outputs: one response per call.

Cost per decision matters. For classify/route workflows, Jev is 7–70× cheaper per token than chat models ($0.042/M vs $0.30–3.00/M).

How to decide

1. Is the output shape complex or nested? Structured Outputs. Jev only handles typed questions with finite options.
2. Is the workflow high-frequency (thousands/minute)? Jev. Chat model cost and latency make high-frequency classify/route impractical.
3. Do you need per-option probabilities and confidence? Jev returns them natively. Structured Outputs: you parse and interpret yourself.
4. Do you need to batch many questions? Jev: ~60 questions in one call. Structured Outputs: one response per call.
5. Is the question which option from a list? Jev. If the model must generate structure: Structured Outputs.

Builds that show the job

Routing, classification, and agent next action — with the cost and latency each builder published on decision APIs.

Duncan

@ephraimduncan

Built a model router with Jev by @typesafeai. Jev decides what model fits your request best and the request is sent to that model.

X·Routing and model choice

A model router on Jev

Gregor Zunic

@gregpr07

Breaking: Browser Use + Jev = Ultrafast ⚡ Findings flights took 7s and cost only $0.0039 🤯 > new action space every step > DOM state space > small LLM fallback to type (this video is at 1x speed btw) Built a tiny open source browser agent. try it below ↓

X·Search

Pick

Flight search with Browser Use

Run time

~7 s

Cost

~$0.004

tamara

@tamarajtran

found the perfect use case for @typesafeai Jev: instant compaction in 2026, why is compaction still a summarization prompt? Jev can make it instant by scoring every tool call and dropping what’s irrelevant

X·Context and memory

Pick

Instant compaction for Claude

Hassan

@nutlope

I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on http://1kpapers.com The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: http://1kpapers.com

Live site·Documents and OCR

1kpapers

Documents

1,018 papers

Total cost

$0.08

Median latency

256 ms

More on Jev use cases. See also Jev vs an LLM for same-job both-ways comparisons.

Try it today

Next steps

• Structured Outputs alternatives — decision APIs (Jev, Clef, Perplexity)
• How to use Jev — your first call, SDKs, gateways
• Jev vs an LLM — same job both ways, with published numbers
• TypeSafe Jev alternatives — Clef, Perplexity, OSS
• Agent prompts — wire Jev into Claude Code, Cursor, Codex

Common questions

What is Structured Outputs?

A feature of chat models (GPT-4o, Claude, etc.) that guarantees the response conforms to a provided JSON schema. Used for structured extraction — parse receipts, label screenshots, extract fields from documents.

How is Jev different from Structured Outputs?

Different category. Jev answers typed questions (Choice/Score/Noul) with finite options, returns per-option probabilities and confidence, costs $0.042/M input. Structured Outputs extracts arbitrary JSON schema from prose, costs $0.30–3.00/M input (chat model pricing).

Can Jev replace Structured Outputs?

No. Different jobs. Jev is for typed questions (which option) in classify/route/act workflows. Structured Outputs is for complex schema extraction from unstructured input. Use Structured Outputs when the output shape is nested or arbitrary; use Jev when the question is which option with confidence.

Which is faster, Jev or Structured Outputs?

Jev: 70–500 ms per TypeSafe docs. Structured Outputs: chat model generation latency, typically seconds for complex schemas. Purpose-built decision models (Jev/Clef/Perplexity) are faster for classify/route/act loops.

Which is cheaper, Jev or Structured Outputs?

Jev: $0.042/M input (output free). Structured Outputs: $0.30–3.00/M input (chat model pricing). For classify/route workflows, Jev is 7–70× cheaper per token.

Do both return probabilities?

Jev returns per-option probabilities and a confidence score for every question. Structured Outputs can return token-level logprobs, but no per-option probabilities for a typed decision. You must parse the JSON and interpret confidence yourself.