Jev 1.14 API
NewLLMThe Jev 1.14 API is TypeSafe AI's typed decision route: one request returns schema-bound choices, scores, and confidence for a state you supply.
- Pricing · per call$0.018 USD
- Context64K
- View tool pricing and billing rules
Use this model through its documented API. An online playground is not available for this protocol yet.
Read the docs01
What this model is for
The Jev 1.14 API is TypeSafe AI's typed decision route: one request returns schema-bound choices, scores, and confidence for a state you supply.
Jev 1.14 is TypeSafe AI's System One Model, served on OmniaKey as the alias jev-latest. You send a state plus a map of typed questions, and the reply gives typed answers instead of prose: a Choice from a closed list, a Score across described levels, or a Noul probability that a statement is true. The catalog lists a 64K window, decision and vision capabilities, and no client tools.
It suits steps where the answer space is small and a program will consume the result: which queue this message belongs to, how severe this report is, whether a claim holds. A generative model is the right tool for the sentence a customer reads. Jev answers every question in one request against the same state, and each answer carries the distribution behind it, so your code decides the threshold. It has its own endpoint, POST https://api.omniakey.com/v1/alpha/search, and takes a native Jev body rather than a chat messages array.
Best-fit coding workloads
Support queue assignment
Ask which team a message belongs to, how annoyed the writer is and whether a human must read it, all in one request. Tip: include an option for cases none of the queues fit, then review the low-confidence answers instead of trusting the top label.
Retrieved passage checks
Score each retrieved chunk for relevance to the question and for contradiction with another chunk before it reaches the prompt of a writing model. Tip: keep source verification in your own code, because a well-scored passage can still be out of date.
Agent action gating
Decide whether a skill applies to the current request, or whether a claim in a tool result is true, before the agent runs anything that changes state. Tip: leave permissions and irreversible checks in code, and treat the probability as one input to that decision.
02
Hands-on notes
Our September 20 smoke test of Jev returned HTTP 200 with typed answers and token usage, and reported jev-1.13.0 as the resolved model.
Read the test
03
Technical specification
Catalog facts and integration details verified for this route.
- Model ID
- jev-latest
- Context
- 64K
- Inputs and tools
- Typed decisions · Vision · Fast responses
- API endpoint
- /v1/chat/completions
- Cache hit / 1M
- Coding clients
Limits to plan around
- There is no free-form output. Jev cannot draft an email, write code or explain itself, so a workflow that needs prose needs a generative model at that step.
- One request holds 64K tokens, and a second limit applies to the state plus the longest single question. Longer material has to be split, or the experimental long-context header has to be enabled at your own risk.
- Text is the default input. Image assets in 1.14 sit behind an experimental header that TypeSafe marks as risky, and audio or video is not part of the interface.
- Exact arithmetic, counting, dates and multi-hop inference are known weak spots. Keep that logic in deterministic code and ask Jev only for the semantic judgement around it.
04
Compare nearby models
Compare workload fit and current listed token rates before choosing a route.
| Model | Best for | Context | Input | Output |
|---|---|---|---|---|
| Jev 1.14This page jev-latest | Bounded semantic judgements that code consumes directly, billed per request rather than per token. | 64K | Official price | Official price |
| GPT-6 Astra gpt-6-astra | Generation, explanation and multi-step reasoning where the written answer is the product. | 1M | $0.9 USD Official price $10 USD | $4.5 USD Official price $50 USD |
| Claude Fable 5.1 claude-fable-5-1 | Long-form analysis and code work that needs composed prose instead of a typed verdict. | 1M | $2.8 USD Official price $10 USD | $14 USD Official price $50 USD |
05
SDK quickstarts
Use one base URL and model ID from cURL, Python, or TypeScript.
curl https://api.omniakey.com/v1/chat/completions \
-H "Authorization: Bearer $OMNIAKEY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "jev-latest",
"messages": [
{"role": "user", "content": "Review this pull request"}
],
"stream": true
}'06
Estimate token cost
Enter expected input and output tokens to compare current listed rates.
- OmniaKey estimate
- $0.00
- Official-rate estimate
- $0.00
- Estimated savings
- $0.00
Estimate covers input and output tokens only. Cache usage, retries, tools, and client-side token accounting can change the final charge.
07
Related coding guides
Deeper selection, setup, and troubleshooting guidance for this model family.
08
Sources and verification
Model facts and links were checked on 2026-10-27. Provider documentation remains the authority for upstream behavior.
Frequently asked questions
How much does Jev 1.14 cost on OmniaKey?
OmniaKey bills $0.018 USD per request rather than per token. The price is the same whether you send a short prompt or use the full context window.
What is the context window for Jev 1.14?
The listed context window is 64K. Actual usable limits can also depend on the client and upstream request format.
What is Jev 1.14 and which version does jev-latest point to?
Jev 1.14 is the current TypeSafe release, version id jev-1.14.0, stable since October 14, 2026, and jev-latest resolves to it. The older 1.13 build can still be called by pinning jev-1.13.0. Record the model field in each response so a version change shows up in your logs rather than in a surprise behaviour shift.
How do I call jev-latest through OmniaKey?
Post the native Jev body to https://api.omniakey.com/v1/alpha/search with your OmniaKey key as a bearer token, and set model to jev-latest. The body holds state and questions, not a messages array, so an OpenAI SDK will not work unchanged. Responses keep your question ids and add a usage block.
How is Jev billed on OmniaKey?
The OmniaKey catalog bills Jev per request rather than per token, and the charge comes out of the account balance. Current figures are in the price table on the model page. Because one request can carry several questions, the cost of a step depends on how many calls you make, not on how long the state is.
What do Choice, Score and Noul return?
A Choice returns the selected option plus a probability for every option and a confidence value. A Score returns a probability-weighted position across the levels you describe, which may be fractional, along with the legend. A Noul returns the probability that the statement is true, and no separate confidence field.
Does Jev 1.14 accept images or Chinese text?
Text is the default: a string, a JSON object or an array of text values. Images need the experimental header in 1.14 and are not something to build on yet. Chinese is accepted, but TypeSafe says English is currently strongest, so test any other language on your own labelled cases before relying on the thresholds.