Jev Can See Now, So I Tried It On My Image Collection

🔍 Read the full analysis: Jev Can See Now, So I Tried It On My Image Collection on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech essentials delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

TypeSafe’s Jev 1.14, released October 14, adds image input and a bounding-box response type to a model previously presented as text-only. In a test of 30 featured images from ThorstenMeyerAI.com, site author Thorsten Meyer reported strong results matching product photos to headlines, but said the broader editorial-image results were less clear; TypeSafe warns that the new features are not ready for production use.

The image feature is available for jev-1.13.0 and jev-1.14.0, but users must explicitly enable it with the header TypeSafe-Danger-Zone-Images: true. TypeSafe also added a box question type in Jev 1.14, which returns one bounding box and a confidence score for a specified target. That feature requires the Jev 1.14 model.

Meyer tested the preview on the 30 most recent images on his site: 10 product roundups, 15 news posts and five essays. For each image, he asked whether it fit its own headline, a randomly selected unrelated post headline, and an off-topic headline about repotting a houseplant. He also asked the model to identify which of 10 products appeared in the product images, running the checks on both Jev 1.13 and 1.14.

For the 10 product images, Meyer said Jev 1.14 identified the matching headline with scores from 0.65 to 0.89, with a median of 0.76. Scores for a randomly paired headline reached no more than 0.18, and the off-topic headline scored no higher than 0.10. Jev 1.14 picked the correct product in nine of 10 cases. Jev 1.13 got all 10 product identifications right, though two answers had confidence below 0.5.

Image Checks Move Into Jev

The release expands Jev from a text-based decision model to a tool that can evaluate visual material through structured questions. That could support repeated tasks such as checking whether an image matches a headline or identifying an object, without requiring a user to ask for open-ended prose. The bounding-box type also offers a structured way to locate a named object in an image.

Meyer’s results suggest a possible use in editorial image review, but they do not establish general reliability. The test was limited to one website’s images and a small set of prompts, and the author’s own account indicates that product matching was clearer than performance on news and essay images. For publishers and developers, the practical point is to run tests on their own material rather than infer performance from this single trial.

TypeSafe’s opt-in design also matters: the image and long-context features are placed behind headers labeled as a “danger zone.” The company says its production reliability is not established. The new capability is available to experiment with, but the vendor’s warning makes it unsuitable to treat the release itself as evidence of production readiness.

From Text Questions to Images

Jev is a decision model that responds to typed questions with outputs such as a yes-or-no probability, a selection from a list, or a score on a scale, each accompanied by confidence. Meyer wrote in September about the model’s approach, which he described as “System One” AI. Before this release, he said, the questions in his use of Jev had to concern text.

TypeSafe’s October 14 release post said image support also works on the existing jev-1.13.0 model, describing that version as having been multimodal already. The new bounding-box type is limited to 1.14. The company also left Jev’s default context limits unchanged at 64,000 tokens per request and 32,000 per question; a separate opt-in header raises both limits to 262,000 tokens. TypeSafe advises against the larger context because reliability can fall on long inputs.

For images, TypeSafe’s request format uses an asset map and a reference string that can be placed in the state or question instructions. The supported still-image formats are PNG and JPEG, with image dimensions from 256 to 1,568 pixels per side. The source account says each file is limited to 16 MiB and each request to 32 MiB; an invalid image can cause the whole request to fail. TypeSafe’s stated input price is $0.042 per million input tokens, with image content billed as ordinary input rather than at a separate image rate.

““There is still way too much jaggedness for us to vouch for these for production use cases.””

— TypeSafe, in its release post as quoted by Thorsten Meyer

Editorial Results Need More Testing

The source account cuts off as Meyer begins discussing results for the 15 news images, so it does not provide the findings for those images or the five essays. The overall outcomes across the full 30-image sample therefore cannot be established from the available report. It is also unclear how the results would change with different websites, image styles, prompts or evaluation criteria.

The test was conducted by the author on images from his own site, and no independent benchmark or third-party replication is described. The report gives confidence scores and selected outcomes, but does not establish how those scores should be calibrated for other users. TypeSafe itself says it cannot yet vouch for production use, while its internal intelligence estimates are not public benchmark scores and, according to Meyer, may be biased toward Jev.

Preview and Model Rollout

Jev 1.14 is available as jev-preview. TypeSafe said jev-latest would move to 1.14 one week after the October 14 release, around October 21, 2026. The company also described deprecation notice periods of 60 days for Jev 1.13.0 and 30 days for 1.14.0, with a formal policy still to come.

For users evaluating the image features, the next practical step is testing with their own images and checking both correctness and confidence rather than relying on the small reported sample. TypeSafe has not supplied a production-readiness date in the source material, and the reliability warning remains in place.

Key Questions

What image features did Jev add?

Jev 1.14 supports image input for structured questions and adds a box type that returns one bounding box and confidence for a target. Image input also works with Jev 1.13.0, according to TypeSafe.

Are the new image features on by default?

No. Users must set the TypeSafe-Danger-Zone-Images: true header to enable image input. TypeSafe says it is not ready to vouch for the feature in production.

How did Jev perform in the reported image test?

In Meyer’s test, Jev 1.14 matched all 10 product images to their own headlines with higher scores than the unrelated comparisons and identified nine of the 10 products. The available report does not include the full results for news and essay images.

Which image formats does Jev support?

The reported limits cover PNG and JPEG still images, between 256 and 1,568 pixels per side. WebP and HEIC images need to be converted first, according to the source account.

When will Jev 1.14 become the latest model?

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

Source: ThorstenMeyerAI.com