I Shared 30 Images With Jev To Try Its New Vision

🔍 Read the full analysis: I Shared 30 Images With Jev To Try Its New Vision on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

TypeSafe released Jev 1.14 on October 14, adding image input and a new bounding-box answer type to its decision model, while warning that the capabilities are not ready for production use. In a test published the next day, Thorsten Meyer of ThorstenMeyerAI.com ran the image feature against 30 images from his site and reported strong product-photo matching, though the available account ends before describing results for news images and essays.

Jev is designed to answer structured questions rather than generate prose: users provide a state and typed questions, and the model returns answers such as yes-or-no probabilities, selections from a list, or scores, with confidence values. Until this release, those questions were text-only. TypeSafe says image input works with both jev-1.13.0 and jev-1.14.0; the new box answer type, which returns one bounding box and a confidence score for a specified target, is limited to Jev 1.14.

The image feature requires the request header TypeSafe-Danger-Zone-Images: true. TypeSafe says images can be placed in the state, question instructions, or criteria used to define answer options. Meyer describes a reference-based setup in which an image is declared once in an assets map and inserted wherever its reference appears. The stated limits are PNG or JPEG still images, 256 to 1,568 pixels per side, up to 16 MiB per image and 32 MiB per request; unsupported formats such as WebP and HEIC must be converted. The source says images are billed as ordinary input tokens at Jev’s normal rate, not with a separate image surcharge.

For his test, Meyer used the 30 most recent featured images on his site: 10 product roundups, 15 news posts and five essays. He asked whether each image fit its own headline, a randomly selected different headline, and an unrelated plant-care headline. For the product images, Jev 1.14 reportedly scored the matching headline from 0.65 to 0.89, with a median of 0.76; random headlines reached no higher than 0.18, and the plant-care headline stayed at or below 0.10. It selected the correct product in nine of 10 cases. The supplied source begins discussing results for news images but is truncated, so it does not provide the remaining findings.

Image Checks Move Into Structured AI

The release extends Jev from text-based decisions to tasks where the input itself is visual. That could let developers test whether an image matches a headline, identify a specified object with a bounding box, or compare a picture against image-based answer choices, while retaining Jev’s structured outputs and confidence scores. Meyer’s product-photo results offer an early example of that kind of use, but they come from a small, site-specific test, not an independent benchmark.

TypeSafe’s opt-in design also matters for teams considering the feature. The company says it has not established production reliability and explicitly warns about remaining problems. Keeping existing behavior unchanged unless a request includes a danger-zone header makes the experimental status visible at the point of use. The release should be read as an invitation to evaluate the feature, not as evidence that image decisions are dependable in high-stakes workflows.

A Text-Only Model Adds Vision

Meyer previously described Jev as a model for structured decisions rather than open-ended writing. Users supply a state and typed questions, such as a yes-or-no check or a choice among named options; the model responds in the requested format with a confidence measure. The October release changes the inputs the system can process, while preserving that decision-oriented approach.

TypeSafe’s release notes also describe a separate long-context option. The default limits remain 64,000 tokens per request and 32,000 per question; an explicit header raises both to 262,000 tokens. The company advises against the larger setting because reliability can fall on long inputs. TypeSafe also says users may opt in to sharing queries to help identify model weaknesses, while stating that it will not train on those queries.

Jev 1.14 is available under the jev-preview name. TypeSafe said jev-latest would move to it about a week after release, around October 21. The source also reports internal intelligence estimates rather than public benchmark scores and notes TypeSafe’s warning that its comparisons against open-weight models may favor Jev. Those estimates do not establish performance on image tasks.

“was multimodal all along”

— TypeSafe, describing Jev 1.13.0 in the release post, according to Meyer

Test Results Stop Mid-Report

The supplied source text is cut off as Meyer begins reporting the results for 15 news images. It does not say how Jev performed on those images or the five essays, nor does it include a complete account of the comparison with Jev 1.13 across the full test. The reported product results therefore cannot establish overall performance across the 30-image set.

TypeSafe has not vouched for the new capabilities in production. The source does not provide an independent evaluation, public vision benchmark, or evidence that confidence scores reliably identify mistakes outside Meyer’s product-image example. It also does not specify how results might vary with different image types, question wording, or operational settings.

Preview Use and Reliability Checks

TypeSafe’s stated rollout plan is for Jev 1.14 to remain available as jev-preview, with jev-latest expected to move to it around October 21, 2026. TypeSafe has described deprecation notice periods of 60 days for Jev 1.13.0 and 30 days for Jev 1.14.0, with a formal policy still to come.

For now, the next meaningful evidence will come from fuller evaluations of the image and bounding-box features. TypeSafe advises users to run their own tests, and the release’s explicit opt-in requirement signals that reliability remains under review. Meyer’s published source, as provided, does not finish its test report, leaving the results for news images and essays unresolved.

Key Questions

What changed in Jev 1.14?

TypeSafe added opt-in image input and a new bounding-box answer type. The box type is limited to Jev 1.14; TypeSafe says image input also works with Jev 1.13.0.

Is Jev’s image feature ready for production?

No production readiness is established in the source. TypeSafe says the capabilities have too many reliability issues for it to vouch for them in production, so image use requires an explicit request header.

How did Jev perform on the product photos?

Meyer reported that Jev 1.14 matched each of 10 product images to its own headline with scores from 0.65 to 0.89, and identified the product correctly in nine of 10 cases. Those figures come from his test, not an independent benchmark.

What happened with the news images and essays?

The source provided for this article is truncated during the news-image results. It does not state the outcomes for the 15 news images or five essays.

When will Jev 1.14 become the latest model?

TypeSafe said Jev 1.14 was available as jev-preview and that jev-latest would move to it about one week after the October 14 release, around October 21, 2026.

Source: ThorstenMeyerAI.com