How I Put Jev’s Visual AI Through A 30-Image Trial

🔍 Read the full analysis: How I Put Jev’s Visual AI Through A 30-Image Trial 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 added opt-in image support and a bounding-box response type to its Jev decision model on October 14, 2026, and AI writer Thorsten Meyer tested the image feature on 30 images from his website. In the results reported so far, Jev 1.14 distinguished product images from competing headlines and identified nine of 10 products correctly, but the account ends before giving the trial’s findings for news images.

Jev is a structured model that responds to typed questions with outputs such as a yes-or-no probability, a choice from a list, or a score with confidence. Before the release, those questions were text-based. TypeSafe’s post says image support also works on jev-1.13.0, as well as the new jev-1.14.0. Images and bounding-box requests require explicit headers, so existing production behavior does not change unless a user opts in.

Meyer used the site’s 30 most recent featured images: 10 product roundups, 15 news posts and five essays. Each image was tested against its own headline, a randomly selected different post headline, and an unrelated headline about repotting a monstera. The 10 product images also went through a multiple-choice task asking Jev to identify the product. Meyer says he ran the questions on both model versions, for about 60 requests overall.

For the product images, Jev 1.14 gave each image’s own headline a score from 0.65 to 0.89, with a median of 0.76. Random alternative headlines scored no higher than 0.18, with a median of 0.06; the plant headline never scored above 0.10. Jev 1.14 named the correct product in nine of 10 cases. Its one error was a mesh Wi-Fi photo it labeled “Wi-Fi 7 router” with 0.60 confidence. The account says all correct answers had confidence of at least 0.77. Jev 1.13 identified all 10 products, though two answers had confidence below 0.5.

A Small Test of Image Judgment

The trial offers an early, task-specific look at whether Jev can apply its structured question format to images, not evidence that the model is reliable across visual work generally. The product results suggest confidence scores may help distinguish stronger answers from a mistake: in this sample, the one product-identification error had lower confidence than every correct answer. That pattern is useful to test, but 10 product photos are a small sample, and it does not establish a dependable threshold for other datasets.

For developers, the release makes image-based classification available without replacing the model’s existing text workflow. The opt-in requirement also signals that TypeSafe considers the capability experimental. The company says the features remain too jagged for it to vouch for production use, so users need to evaluate results on their own images and tasks before relying on them.

Jev’s Opt-In Image Release

TypeSafe published Jev 1.14 on October 14, 2026. Its release adds image input, the new “box” type for returning one bounding box and confidence, and a separate option to raise the context limit. Image requests require the TypeSafe-Danger-Zone-Images header; the box type is limited to Jev 1.14. TypeSafe also kept image support available on Jev 1.13, which its post described as having been multimodal already.

The feature has practical constraints: the supplied material says images must be PNG or JPEG stills, between 256 and 1,568 pixels per side, with a 16 MiB per-file limit and 32 MiB per request. A failed image can fail the whole request. TypeSafe recommends multipart uploads, while Base64 encoding adds size overhead. The company also advises against the longer-context option because it says reliability declines on long inputs.

TypeSafe shared internal intelligence estimates rather than public benchmark scores and cautioned that its comparisons, limited to open-weight models, may favor Jev. Meyer said he chose to run his own evaluation instead. His trial is an individual test on one site’s images, not an independent benchmark or a company performance claim.

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

— TypeSafe, in its Jev 1.14 release post

News-Image Results Are Missing

The supplied account stops as Meyer begins discussing the 15 news images, so it does not provide their scores or explain how Jev handled those headlines. It also gives no results for the five essays. The available findings therefore cover the 10 product images in detail, but not the full 30-image set described in the test plan.

The material does not report enough information to judge statistical reliability, including repeated trials, image-selection criteria beyond recency and category, or whether the questions and scoring method were fixed before testing. The results are one author’s evaluation, not a controlled benchmark. TypeSafe’s warning about production reliability remains in place.

More Evaluation Before Deployment

TypeSafe made Jev 1.14 available as jev-preview and said the jev-latest alias would move to it about one week after release, around October 21, 2026. The source material does not confirm whether that change has since occurred. The company said it would give 60 days’ deprecation notice for Jev 1.13 and 30 days for Jev 1.14, with a formal policy still to come.

For now, the practical next step is testing the opt-in feature on the images and questions a team actually uses, while treating confidence as a signal to evaluate rather than proof of correctness. A fuller account of Meyer’s news-image and essay results would clarify whether the product-photo pattern held across the rest of the trial.

Key Questions

What changed in Jev 1.14?

TypeSafe added opt-in image input and a new box response type that returns one bounding box and a confidence. The box type is available on Jev 1.14; TypeSafe says image support also works on Jev 1.13.

How did Jev perform on the product images?

In Meyer’s 10-image product test, Jev 1.14 identified nine products correctly. It also scored each image’s own headline higher than the alternative and unrelated headlines in the reported results.

Did Jev 1.13 do better than Jev 1.14?

On product identification alone, Jev 1.13 got all 10 answers right, while Jev 1.14 got nine right. Meyer reported that two of Jev 1.13’s correct answers had confidence below 0.5; the supplied account does not provide a complete comparison across all image categories.

Is TypeSafe recommending Jev images for production?

No. TypeSafe said the new capabilities remained too jagged for it to vouch for production use and placed them behind explicit opt-in headers.

What are the trial’s main limitations?

The reported product findings come from just 10 images on one website. The supplied account ends before news-image results and does not give results for the five essays, so it cannot establish performance across the full 30-image sample.

Source: ThorstenMeyerAI.com