TypeSafe · 14 Oct 2026 · jev-1.14.0

Jev 1.14 sees images, draws boxes and reads long context

TypeSafe’s newest decision model adds three experimental features: PNG/JPEG image input, a fourth question type — Box — that returns a bounding box, and long context up to 262k tokens. Same typed answers, same price per input token. Try each one below with a single click.

Try Jev 1.14 now Read the API docs

Image input

Ask typed questions about a picture

Attach up to 4 PNG or JPEG images and ask choice, yes/no or score questions about screenshots, product photos or camera frames. Images are billed as input tokens.

Box

Locate one thing in an image

A Box question returns top, bottom, left and right as fractions of the image plus a confidence score. The playground draws the box on your image.

Long context

Decide over a whole document

Send a full log, contract or conversation — up to 262k tokens. jevmodel.org switches on the long-context opt-in for you when a request needs it.

Live playground

Run Jev 1.14 on real requests

These presets run on jev-preview, which points to jev-1.14.0. Swap in your own image or text. Guests get trial tokens; long context needs a free account.

Model

1 Text to judge

0 / 8,000 guest limit

2 Questions

0 / 8

10,000 input tokens without signup.
100,000 when you sign in.

Request preview

Your draft stays in this browser tab.

Your judgment

Real Jev output

A question in. A judgment out.

Your result appears here with the response returned by Jev. Run the example to see how the question and answer options shape the judgment.

Import configuration

Model judgment, not a guarantee. Review the result before using it to trigger an action.

What changed

Jev 1.13 vs Jev 1.14

	Jev 1.13	Jev 1.14
Model ID	jev-1.13.0	jev-1.14.0
Alias	jev-latest (stable)	jev-preview
Input	Text and JSON	Text, JSON and PNG/JPEG images (experimental)
Question types	choice · noul · score	choice · noul · score · box (experimental)
Default context	64k per request · 32k per question	64k per request · 32k per question
Long context (opt-in)	Up to 262k tokens	Up to 262k tokens
Price	$0.042 / 1M input tokens, output free	Same — images count as input tokens

API

Call it from your code

Same endpoint and key as before. Set model to jev-preview (or pin jev-1.14.0). jevmodel.org adds TypeSafe’s opt-in headers for images, Box and long context automatically.

Image + Box in one request

curl https://jevmodel.org/v1/systemone \
  -H "Authorization: Bearer $JEVMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-preview",
    "state": "Bug report screenshot: shot-1",
    "assets": {
      "shot-1": { "type": "image", "format": "png", "data_base64": "<base64>" }
    },
    "questions": {
      "blocking_error": { "type": "noul", "instructions": "Does the screenshot show a blocking error?" },
      "pay_button": { "type": "box", "instructions": "the Pay button in the screenshot" }
    }
  }'

# → "pay_button": { "type": "box",
#     "box": { "top": 0.81, "bottom": 0.875, "left": 0.08, "right": 0.575 },
#     "confidence": 0.9 }

Long context: just send the full text

curl https://jevmodel.org/v1/systemone \
  -H "Authorization: Bearer $JEVMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<'JSON'
{
  "model": "jev-preview",
  "state": "<the whole incident log, contract or thread — up to 1,000,000 characters>",
  "questions": {
    "root_cause": { "type": "choice",
      "instructions": "What caused the checkout failures?",
      "criteria": { "database": "...", "deploy": "...", "payment_provider": "...", "traffic": "..." } }
  }
}
JSON

# Above ~30k input tokens jevmodel.org adds
# TypeSafe-Danger-Zone-Long-Context: true for you (262k max).

Before you ship

Experimental means it can change

Images, Box and long context are experimental (“Dreamer-ready”). Request format, answers, limits and accuracy may change between releases.
jev-latest still points to jev-1.13.0. Use jev-preview or pin jev-1.14.0 to get the new features.
A Box answer always contains a box — there is no “not found”. Pair it with a yes/no question when the target may be missing.
Each Box question finds one target. Use one question per target, and crop large images when the target is small.
Long context is slower and less accurate than staying inside 64k / 32k. Trim what you can.
On jevmodel.org: up to 4 images per request, 256–1568 px per side, about 4 MB each. Guests are limited to 8,000 characters of text.

FAQ

Is Jev 1.14 the default model now?

No. jev-latest still points to the stable jev-1.13.0. Jev 1.14 is reachable as jev-preview or by pinning jev-1.14.0.

Can Jev 1.14 generate or edit images?

No. It reads images and returns typed answers — choice, yes/no, score or a bounding box. It does not generate text or images.

How are images billed?

As input tokens, at the same per-token rate. The usage block of each response shows the real count, and jevmodel.org charges exactly that.

Do I need special headers?

Not on jevmodel.org. TypeSafe requires opt-in headers for images and long context; this site adds them for you when a request needs them.