Jev’s AI Vision In Action: My Test With 30 Images

🔍 Read the full analysis: Jev’s AI Vision In Action: My Test With 30 Images on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets 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 released Jev 1.14 on October 14, adding opt-in image input and a new bounding-box response type to its decision model. In a test published the next day, Thorsten Meyer of ThorstenMeyerAI.com said Jev distinguished product images from mismatched headlines in a sample of 30 website images. TypeSafe said the new capabilities remain too unreliable for production use without further evaluation.

Meyer tested the 30 most recent featured images on his site: 10 product roundups, 15 news posts and five essays. Each image was assessed against its own headline, a randomly selected headline from another post, and an unrelated headline about repotting a plant. Product images also received a multiple-choice question asking which of 10 products each picture showed. Meyer says he ran the questions on both jev-1.14.0 and jev-1.13.0, making about 60 requests in total.

For the 10 product images, Meyer reported that Jev 1.14 separated matching headlines from mismatches: the images’ own headlines received scores from 0.65 to 0.89, with a median of 0.76. Random other headlines scored no higher than 0.18, with a median of 0.06, while the plant headline stayed at or below 0.10. Jev 1.14 identified the product correctly in nine of 10 cases. Meyer said the incorrect answer had a confidence score of 0.60; the correct answers were all at least 0.77.

The supplied source begins a section on results for news images but ends mid-sentence, before giving those findings. It therefore does not provide the complete results for the 15 news images or five essays, nor enough information to independently evaluate the test. The figures are Meyer’s account of a small, site-specific experiment, not a public benchmark or independent assessment.

What Image Decisions Could Enable

Jev’s image input extends a model designed for structured decisions beyond text-only tasks. A user can ask a question with a yes-or-no answer, request a choice from defined options, or have the system assign a score. Meyer’s test examines whether the system can assess whether a featured image corresponds to a post’s headline, a task that could be applied to website publishing workflows.

Meyer reported that Jev separated relevant product photos from unrelated headlines in his sample. The test does not establish whether the system can reliably assess image quality, verify factual accuracy, or handle visuals from other sites. TypeSafe has said the image features are not yet reliable enough for production use. Any team considering automated editorial checks would need to evaluate the system on its own material.

The Opt-In Jev 1.14 Release

Before this release, Jev’s typed questions applied to text. TypeSafe’s October 14 post said image support also works on jev-1.13.0, describing that earlier model as multimodal “all along.” The company added a new box type in Jev 1.14, which returns one bounding box and a confidence score for a described target. Image support and bounding boxes are separate additions; the box type is specific to Jev 1.14.

TypeSafe put the features behind explicit request headers: TypeSafe-Danger-Zone-Images: true enables image input, while a separate header enables longer context. The company said existing production behavior would not change unless callers opt in, and cautioned that it could not vouch for the new capabilities in production use. Image files must be PNG or JPEG stills, 256 to 1,568 pixels per side, with a maximum of 16 MiB per image and 32 MiB per request, according to Meyer’s account of the release details. WebP and HEIC files need conversion first.

Jev 1.14 was offered as jev-preview at release, with jev-latest scheduled to move to it around October 21, one week later. TypeSafe also retained its prior default context limits rather than automatically raising them, citing reliability concerns on longer inputs. Meyer said the company shared internal intelligence estimates instead of public benchmark scores and advised readers to run their own evaluations.

“jev-1.13.0 was multimodal all along.”

— TypeSafe, in its release post as described by Meyer

The Missing News-Image Results

The supplied article text stops at the beginning of Meyer’s discussion of the 15 news-post images. It does not say how Jev scored those images, what happened with the five essays, or whether the two model versions differed on those groups. Those results cannot be inferred from the product-photo findings.

The test was conducted on one site’s images and headlines, and the source provides no independent replication or public benchmark. It also does not explain enough about repeated runs, scoring calibration, or the selection of the random headlines to establish how broadly the results apply. TypeSafe’s stated production warning remains in force, and the source does not report a separate TypeSafe validation of Meyer’s findings.

Evaluations Before Wider Use

TypeSafe’s immediate rollout plan was to offer Jev 1.14 through jev-preview, then move jev-latest to the new version around October 21, 2026. The company said it would give 60 days’ deprecation notice for version 1.13.0 and 30 days for 1.14.0, while a formal deprecation policy was still to come.

Users considering image tasks can evaluate the feature on their own images and set confidence thresholds for their workflows. The supplied source does not include the full results of Meyer’s test, so it does not establish how Jev performed on news images. TypeSafe has cautioned that the features are not yet ready for production use.

Key Questions

What changed in Jev 1.14?

Jev 1.14 added image input support and a box response type that returns one bounding box with a confidence score. Image input also works on Jev 1.13.0, according to TypeSafe’s release post as reported by Meyer.

What did Meyer’s image test find?

For 10 product images, Meyer reported that Jev 1.14 scored each image’s own headline higher than unrelated headlines and selected the correct product in nine of 10 cases. The source supplied for this article cuts off before the news-image findings.

Is Jev’s image support ready for production?

TypeSafe says it cannot yet vouch for the features in production use cases because of reliability issues. The image and bounding-box capabilities require an explicit opt-in header.

Which image files does Jev accept?

According to Meyer’s account of the release specifications, Jev accepts PNG and JPEG still images from 256 to 1,568 pixels per side. Each file can be up to 16 MiB, and a request can contain up to 32 MiB; WebP and HEIC files must be converted first.

Source: ThorstenMeyerAI.com