Jev version guide · documentation read October 7, 2026
Jev 1.13.0
The current Jev model: what TypeSafe publishes about it, where it stands on JevBench v1.6.0, and how to call it.
jev-1.13.0 is the release behind the aliases jev-latest and jev-preview. It is a hosted, text-only System One model with a 64k-token context that answers typed questions in parallel and returns calibrated probabilities. This page gathers every published number about the version in one place, with the date it was read, so you can pin, budget and rate-limit against the right figures.
Version and aliases
From TypeSafe’s model documentation and the OpenRouter listing.
- Current version
- jev-1.13.0
- Aliases
- jev-latest and jev-preview both resolve to jev-1.13.0; TypeSafe currently publishes no separate preview build
- Dated snapshot
- typesafe/jev-1.13-20260917 on OpenRouter, a 17 September 2026 build of the same version
- Developer
- TypeSafe
- Model class
- System One: typed decisions, no text generation
- Post-training
- RLCD, reinforcement learning for calibrated decisions
Context window, inputs and limits
What a single request can carry, and what the model is documented to do with it.
- Context window
- 64k tokens per request; 32k for the state plus the longest single question
- Inputs
- Text only: a string, a JSON object or an array of text values. No images, audio or video
- Question types
- Noul (yes/no probability), choice (one of your options, with every probability), score (ordered levels you define)
- Parallelism
- The state is read once and every question is evaluated against it in parallel
- Languages
- English is the primary training language and where accuracy is best; other languages, including CJK scripts, are handled but not equally well
- Customer data
- TypeSafe states that Jev is not trained on customer requests or responses
Pricing and rate limits
TypeSafe’s published figures and how the same model is billed on Jev AI.
- TypeSafe list price
- $0.042 per million input tokens; output tokens are free
- TypeSafe rate limits
- 100,000 tokens per second and 80 requests per second as published on October 7, 2026; TypeSafe adjusts them dynamically
- On Jev AI, playground
- 1 credit per run, any length
- On Jev AI, API
- Metered on input tokens: 1 credit per 1,000,000 input tokens, accumulated across requests
- Welcome credits
- 20 on a new account
TypeSafe states that its rate limits adjust dynamically; the figures above are the ones published on October 7, 2026. Jev AI plans and credit packs are on the pricing page.
Jev 1.13.0 on JevBench v1.6.0
Benchmark Heaven measured the hosted API on 600 decisions, a fresh sealed draw plus the public set. Read on October 6, 2026.
| Official rank | #2 of 92, composite 71.1; Quyet-1.0-Large (71.4) ahead |
| Capability ranking | #3 at 76.5, the mean of intelligence and calibration among Jev-class systems |
| Intelligence | 62.4 — open half 65.6, sealed half 63.6 |
| Calibration | 90.6 |
| Speed | 91.5, from a 0.24s median / 0.30s p95 over the network |
| By request type, chance-corrected | Choice 76.0 · yes/no 57.0 · score 60.8 |
The full top ten, both rankings and the method are on the JevBench results page. Head-to-head pages cover the open alternatives measured on the same pool: Cygnet, Winnow-12B, decider-4b, JevK5 and Imajev-4B.
Pin the version in a request
Name jev-1.13.0 explicitly when you want the answer to stay stable across a future release; use jev-latest when you want upgrades automatically.
curl --fail-with-body https://jev-ai.pro/api/v1/systemone \
-H "Authorization: Bearer $JEV_AI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "jev-1.13.0",
"state": "Order 8841: the customer asks to cancel and wants a refund for the unused month.",
"questions": {
"intent": { "type": "choice", "instructions": "What is the customer asking for?",
"criteria": { "cancel": null, "refund": null, "both": null, "other": null } },
"urgent": { "type": "noul", "instructions": "Does this need a reply today?" }
}
}'Create the key on the Jev API page. The official TypeSafe SDK works against this endpoint with its base URL set to https://jev-ai.pro/api; the documentation has the full request and response reference and the error codes.
What 1.13 is documented to be weak at
TypeSafe publishes its own list; designing around it is most of the craft.
- Literal reading. Jev answers the question you wrote. Put boundary cases in the criteria.
- Arithmetic, counting and dates. Keep them in code; use Jev to extract or classify the pieces.
- Long, noisy state. Accuracy falls when the state is full of unrelated text. Filter before sending, even within the 64k limit.
- Adversarial text. Injected instructions inside the state can move answers. Test edge cases; the prompt-injection judge is one guard.
- Generation. Jev does not write. Pair it with a generative model when you need text.
Jev 1.13.0 FAQ
What is the latest Jev version?
jev-1.13.0. Both aliases that TypeSafe documents, jev-latest and jev-preview, resolve to it as of October 7, 2026, so naming jev-1.13.0 explicitly is the only way to pin the version if a newer release ships.
What is Jev’s context window?
64k tokens per request. The state can use up to 32k of that, plus the longest single question; TypeSafe’s own guidance is to filter long, noisy inputs before sending them, because accuracy falls when the state is full of unrelated text.
What are Jev’s rate limits?
TypeSafe publishes 100,000 tokens per second and 80 requests per second as of October 7, 2026, and notes that the limits adjust dynamically with capacity. Earlier in September the published figures were higher, so check the documentation rather than hard-coding a number.
How much does Jev 1.13 cost?
TypeSafe lists $0.042 per million input tokens with free output tokens. On Jev AI the same model costs 1 credit per playground run or 1 credit per 1,000,000 input tokens through the API, with 20 welcome credits on a new account.
How does Jev 1.13.0 score on JevBench?
JevBench v1.6.0 ranks it #2 of 92 with a composite of 71.1: intelligence 62.4, calibration 90.6, speed 91.5, and a 0.24s median / 0.30s p95 latency on the hosted API. On Benchmark Heaven’s Capability ranking it is #3 at 76.5.
Is there a Jev 1.13 model card or changelog?
TypeSafe’s documentation describes the model, its limits and its pricing, but publishes no release notes or changelog for 1.13.0. The dated OpenRouter snapshot, typesafe/jev-1.13-20260917, is the only public record of when the build shipped.
Can Jev 1.13 read images?
No. It accepts text only. Jev-Omni, OmniJev, Imajev-4B and Clef are the image-capable decision models this site documents.
Is Jev 1.13 open source?
No. Jev is a proprietary hosted model and TypeSafe does not disclose its size or architecture. The open alternatives measured on the same benchmark have their own guides on this site.
Sources
Read on October 7, 2026; figures can change.
Jev is developed by TypeSafe. Jev AI is an independent playground and API; accounts, keys and balances here belong to Jev AI.