Overview
Jev 1.14 is the second release of TypeSafe AI's System One decision model, announced on the TypeSafe blog on 14 October 2026 under the model id jev-1.14.0. Like Jev 1.13 it does not generate prose: a request carries a state plus typed questions, and the answer comes back as a typed value with probabilities. TypeSafe describes the release as "small, subtle improvements to everything", with the largest expected gain in multilingual support.
The release shipped first behind the jev-preview alias, which TypeSafe defines as "the most recent release, official or not", while jev-latest — "the most recent stable, official release" and the SDK default — stayed on jev-1.13.0, with the announcement stating that jev-latest would move to 1.14.0 one week later. TypeSafe kept Jev 1.13.0 available and promised 60 days' warning before deprecating it, and 30 days' warning for 1.14.0. Pricing is unchanged: $0.042 per million input tokens, with output tokens free.
Jev 1.14 also arrived with a set of experimental "danger zone" features, each switched on by an API header and, in TypeSafe's words, still carrying "way too much jaggedness" for production. Image input lets PNG or JPEG images sit in a request's assets map and be referenced from the state, instructions or criteria. A new Box question type, available only on Jev 1.14, returns a single bounding box as four edges plus a confidence. A long-context mode raises both the per-request and per-question limits to 262K tokens, at the same per-token price but slower and less accurate. TypeSafe's post mentions internal intelligence measurements but publishes no benchmark numbers, so this page carries none.
| Released | 2026-10-14 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Context | 64K |
| Max output | A typed value with probabilities and a confidence — no generated text |
| Modalities | Text, Image (experimental) |
| Status | Preview |
Pricing
| Input | $0.042 / 1M tokens |
|---|---|
| Output | $0.00 / 1M tokens |
Charged per input token; output tokens are free. TypeSafe also quotes the input price as $42 per billion tokens. Long context costs the same per token.
Strengths
- Improved multilingual support over Jev 1.13, per TypeSafe's release post
- Box questions return one bounding box as top, bottom, left and right edges plus a confidence — exclusive to Jev 1.14
- Experimental PNG and JPEG image input placed anywhere in the state, instructions or criteria
- Optional long-context mode raises both context limits to 262K tokens
- Same price as Jev 1.13: $0.042 per 1M input tokens, output tokens free
Best for
- Routing, classification and triage decisions inside software, now including non-English inputs
- Asking typed questions about screenshots, scanned forms or other still images
- Locating a single element in an image, such as a field on a form or a button on a screen
- Decisions over long documents when the long-context header is enabled
How to access
| Provider | Model ID |
|---|---|
| TypeSafe AI ↗ | jev-1.14.0 |
Jev (System One) — every version
The full lineage of the Jev (System One) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Jev 1.14 | 2026-10-14 | 64K | Proprietary |
| Jev 1.13current | 2026-09-15 | 64K | Proprietary |
FAQ
Which model id do I call for Jev 1.14?
jev-1.14.0, or the jev-preview alias, which TypeSafe points at the most recent release whether official or not. The announcement said the stable jev-latest alias, the default in TypeSafe's SDKs, would move from jev-1.13.0 to 1.14.0 one week after release.
How do I send images or Box questions?
Send the header TypeSafe-Danger-Zone-Images: true with each request that contains an image or a Box question; without it the API returns 400. Images are PNG or JPEG, 256 to 1568 pixels on each side and at most 16 MiB, listed once in the assets map and referenced where they belong.
How long can a Jev 1.14 request be?
By default 64K tokens for the state plus all questions, and 32K for the state plus one question. With the TypeSafe-Danger-Zone-Long-Context: true header both limits rise to 262K tokens, at the same per-token price but slower and less accurate.
What are the limits of Box questions?
Each Box answer is a single box, so several targets need one question each. There is no "not found" answer — a Box question about something absent still returns four edges — so TypeSafe suggests a separate Noul question to check presence, and cropping first when the target is small in a large image.