How I Tested Jev’s Image Skills On 30 Personal Photos

🔍 Read the full analysis: How I Tested Jev’s Image Skills On 30 Personal Photos on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys 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

Thorsten Meyer tested TypeSafe’s Jev 1.14 on 30 featured images from his site after the company added opt-in image support and a bounding-box response type on October 14, 2026. In the product-photo portion of his test, the model distinguished images from their own headlines and named the pictured product in 9 of 10 cases, though TypeSafe has warned that the new capabilities are not yet ready for production use.

Meyer selected 30 recent featured images from ThorstenMeyerAI.com: 10 for product roundups, 15 for news articles and five for essays. He asked the model whether each image fit its own headline, a randomly selected different headline and an unrelated headline about repotting a houseplant. He also asked Jev to identify which of 10 products appeared in each product image. The test used Jev 1.14.0 and 1.13.0 across about 60 requests.

For the 10 product images, Meyer reported that Jev 1.14 gave each image’s own headline a score 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, and the plant headline scored no higher than 0.10. These are results from one small test set, not a general benchmark of image understanding.

On product identification, Jev 1.14 named the correct item in 9 of 10 images. The error involved a mesh Wi-Fi photo showing a router with antennas alongside a cylindrical mesh unit; Jev called it a “Wi-Fi 7 router,” with confidence of 0.60. Meyer said the nine correct answers had confidence scores 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 Matching

The test offers an early, task-specific look at whether a model built around structured decisions can use images as input. Meyer’s results suggest Jev 1.14 may help flag an image that appears poorly matched to a headline or classify items in a limited product set. That could matter for publishers and other teams handling many images, where manual checks take time.

But the findings do not establish that Jev is dependable across image types, subjects or real-world workflows. The sample was 30 images from one website, and the reported product results cover only 10. The distinction between a useful screening signal and a reliable production decision matters, particularly when a wrong classification could affect a published article or downstream process.

Jev Adds Opt-In Image Input

Jev is a decision model that responds to typed questions with outputs such as yes-or-no probabilities, selections from a list or scores on a scale, each paired with a confidence value. Before this release, those questions were text-based. TypeSafe’s October 14 announcement said image support also works with jev-1.13.0, describing the earlier model as multimodal even though the feature had not been presented for use.

Images require the request header TypeSafe-Danger-Zone-Images: true. TypeSafe also introduced a “box” question type in Jev 1.14, which returns one bounding box and a confidence score for a described target. The company said the features retain explicit opt-in requirements because it sees too much “jaggedness” to vouch for them in production.

Image files must be PNG or JPEG stills, between 256 and 1,568 pixels per side, with limits of 16 MiB per file and 32 MiB per request. TypeSafe says images are billed as ordinary input tokens at its normal rate, with no image surcharge. The company also advises against using its optional longer context limit because reliability can fall on long inputs.

““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

Limits of the 30-Image Test

The supplied account cuts off as Meyer begins discussing the 15 news images, so their results and the findings for the five essays are not available here. It is also unclear whether the roughly 60 requests included repeated runs, how the images and headlines were selected beyond the stated categories, or whether the questions were evaluated against independent human judgments.

The test does not show how Jev performs on a larger or more varied collection, or whether its confidence scores are calibrated across tasks. TypeSafe has not endorsed the image features for production, and the article does not provide a separate benchmark or an independent replication. The results should be read as a personal experiment, not proof of broad accuracy.

Release Rollout and Further Testing

TypeSafe made Jev 1.14.0 available as jev-preview and said jev-latest would move to it about a week after the October 14 release, around October 21. The company described the model update as a set of small, broad improvements, with multilingual support expected to benefit most. It also said it would give 60 days’ notice before deprecating version 1.13.0 and 30 days for version 1.14.0, ahead of a formal policy.

For users considering image tasks, the next useful evidence would be larger tests that report outcomes across image categories, repeatability and confidence performance. Meyer’s account points toward that need but does not provide the missing news-image and essay results in the supplied text. TypeSafe continues to characterize the new capabilities as experimental, so production readiness remains unresolved.

Key Questions

What did Jev add in version 1.14?

TypeSafe added opt-in image input and a new “box” response type that returns one bounding box with a confidence score. The image capability also works with Jev 1.13.0, according to the company.

How many images were included in Meyer’s test?

Meyer tested 30 featured images: 10 product-roundup images, 15 news images and five essay images. The supplied source gives detailed results only for the product-image portion.

How did Jev 1.14 perform on product images?

In Meyer’s test, Jev 1.14 matched each of the 10 product images more strongly to its own headline than to the comparison headlines, and identified the pictured product correctly in 9 of 10 cases. These findings apply to this small, single-site test.

Is TypeSafe recommending Jev image support for production?

No. TypeSafe said there was still too much “jaggedness” to vouch for the new capabilities in production use cases and placed image support behind an explicit opt-in header.

What remains unknown about the test?

The supplied account does not include results for the 15 news images or five essays. It also does not establish how well the findings generalize to other image collections or whether the confidence scores are reliable across tasks.

Source: ThorstenMeyerAI.com