Jev 1.14: What's New – Images, Box Type & Long Context

01 / WHAT CHANGED

A model update, plus opt-in experiments

TypeSafe AI describes improvements across Jev's existing tasks and expects the largest gain in multilingual support. That is the vendor's assessment, not a Jev Playground benchmark or a guarantee for a particular language.

Jev remains a System One Model for software. The release also exposes experimental images, a new Box Question, and longer context. Image support extends to Jev 1.13 too; Box requires Jev 1.14.

This guide checks documentation, not model quality. No new Jev 1.14 evaluations were run for this page.

02 / MODEL IDS & ALIASES

Select a version deliberately

As checked October 19, 2026, TypeSafe's model documentation still maps jev-latest to jev-1.13.0 and jev-preview to jev-1.14.0. The launch post planned to move jev-latest roughly a week after release; the documented alias has not yet moved.

Jev 1.13

Versioned ID: jev-1.13.0
Documented alias: jev-latest
Experimental features: Images and longer context; no Box Questions.

Jev 1.14

Versioned ID: jev-1.14.0
Documented alias: jev-preview
Experimental features: Images, Box Questions, and longer context.

Aliases can move without a code change. For a controlled upgrade, use a versioned ID in your own integration and record the returned model version.

Experimental images and Box

03 / DANGER ZONE

Images and a new bounding-box type

Experimental, not production-ready.

TypeSafe AI explicitly cautions against relying on these danger-zone capabilities in production. A valid response shape does not establish that an image judgment or location is correct.

Images on Jev 1.13 and 1.14

Send TypeSafe-Danger-Zone-Images: true on your own TypeSafe API request. Define images in the top-level assets map, then reference them in state, a Question's instructions, or its criteria. The API replaces those references with the images.

TypeSafe AI says Jev 1.13 already had multimodal capability and warns that image tasks perform worse than text-only tasks. Image support is therefore not exclusive to 1.14. Native audio and video input remain unsupported.

Box Questions on Jev 1.14 only

A box Question locates a target in an image and returns one bounding box. Use jev-1.14.0 or jev-preview; Box also requires the images header and an image in the request. It is a localization result, not generated prose or an automatic action.

The official Images and Box documents are linked in Sources. This site does not add these request fields or headers for you.

Default and experimental context limits

04 / CONTEXT LIMITS

Two limits, with a larger experimental option

Both versions default to 64k tokens per request and 32k per Question. The request limit covers the State plus all Questions; the Question limit covers the State plus any one Question.

For either version, TypeSafe-Danger-Zone-Long-Context: true raises both limits to 262k tokens. TypeSafe documents unchanged per-token pricing, but slower and less accurate results than within the default limits. A larger State also consumes more input tokens.

This API option does not extend Jev Playground's separate 8,000-character text limit. Characters and model tokens are different units.

05 / PRICING

The published token price is unchanged

TypeSafe lists the same rate for Jev 1.13 and 1.14: $0.042 per million input tokens, or $42.00 per billion. Output tokens are $0.00 at the published rate.

The rate is an API price, not a charge to use this website. Use the existing pricing calculator to estimate a text workload. Do not assume that longer inputs cost the same per request just because the per-token rate is unchanged.

Upgrade and data-sharing considerations

06 / UPGRADING FROM 1.13

Check your workflow before moving an alias

Evaluate your own representative cases against a versioned model before changing your integration. Review routing thresholds and failure handling; an expected improvement does not guarantee identical probabilities or decisions.

TypeSafe plans to keep Jev 1.13 available for a while. Its release post expresses an intention to give 60 days' notice before formally deprecating 1.13 and 30 days for 1.14, with a formal deprecation policy still to follow. These are announced intentions, not a published retirement date.

TypeSafe also announced optional query sharing to help investigate model issues, while saying those queries will still not be used for training. This opt-in is separate from model selection; consult the provider's data terms before sharing sensitive material.

07 / TRY JEV ONLINE

Try the existing text-based Playground

You can try Jev's Boolean, Choice, and Score decisions here for free, without signup or your own API key. Usage limits and availability apply.

Try Jev Online Jev API Guide

This is not a Jev 1.14 demo.

Jev Playground has no model version selector: you cannot select Jev 1.14 or jev-preview here. It accepts text-only State up to 8,000 characters, does not support images or Box Questions, and does not enable experimental long context or send danger-zone headers.

The production site uses Vercel AI Gateway with typesafe-ai/jev, not a visitor-selected TypeSafe version. The Direct adapter uses jev-latest when configured. This guide does not identify the Gateway model slug with a particular version.

To select 1.14 or use the experimental features, use your own TypeSafe API key and the required request headers in your own integration. Start with the API guide and the official sources below. This website does not issue API keys.

Questions and answers

08 / QUICK ANSWERS

Jev 1.14 FAQ

Does jev-latest already mean Jev 1.14?

Not in the model documentation checked October 19, 2026: it still points to jev-1.13.0. Use jev-preview or jev-1.14.0 to select 1.14 in your own TypeSafe integration.

Are images exclusive to Jev 1.14?

No. The experimental images header also enables images on Jev 1.13. Box Questions require Jev 1.14.

Do I need the experimental headers for ordinary text?

No. The default text interface and context limits remain available without danger-zone headers. Enable only the experiments your own request needs.

Can I test images or Box in this Playground?

No. This site exposes text-based Boolean, Choice, and Score tools, with no version selection, image input, Box type, or experimental long context.

Does Jev 1.14 cost more?

The current TypeSafe model table lists the same input rate for both releases and free output tokens. Actual input volume still affects cost.

Sources

09 / SOURCES & VERIFICATION

Official release and model documentation

Official sources checked October 19, 2026. Jev Playground is independent of TypeSafe AI. Release claims are attributed to the vendor; model aliases and experimental interfaces can change.

Jev 1.14 in 30 seconds