Testing Jev’s AI Vision With Images From My Own Collection

🔍 Read the full analysis: Testing Jev’s AI Vision With Images From My Own Collection 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 image input and a new bounding-box response type to its Jev decision model in Jev 1.14, released October 14, 2026. In a test published the next day, Thorsten Meyer of ThorstenMeyerAI.com ran the model on 30 of his site’s featured images; the reported product-photo results were strong, but the available account ends before giving the findings for news images.

Jev is designed to return structured judgments rather than write prose: users provide a state and typed questions, and the model responds with probabilities, selections or scores, along with confidence values. Before the release, those questions were text-based. TypeSafe says image references can now appear in the request’s state, question instructions or criteria. The image capability is available on jev-1.13.0 and jev-1.14.0, but requires the explicit header TypeSafe-Danger-Zone-Images: true.

Jev 1.14 also introduces a box type that returns one bounding box for a described target, with a confidence score. That feature is limited to Jev 1.14 and is likewise presented as experimental. TypeSafe said the new capabilities have too much “jaggedness” for it to endorse them for production use, and placed them behind opt-in headers. The release also includes a separate header for raising the request and question context limits from 64,000 and 32,000 tokens to 262,000; TypeSafe cautions that reliability can fall on long inputs.

Meyer tested 30 recent featured images: 10 for product roundups, 15 for news posts and five for essays. He asked whether each image fit its own headline, a randomly chosen different headline, and an unrelated headline about repotting a Monstera plant. For the 10 product images, Jev 1.14’s scores for the correct headline ranged from 0.65 to 0.89, with a median of 0.76. Random headlines scored no higher than 0.18, with a median of 0.06; the plant headline scored no higher than 0.10. On a separate product-identification question, Jev 1.14 selected the right product in nine of 10 cases.

Where Image Checks Could Help

The test addresses a practical use for a model built around repeatable, narrow decisions: checking whether an image matches a headline. Meyer said he chose the task because featured images are assigned to every post on his site, but checking that match is not routinely done by hand. If the results hold across broader tests, image input could let developers add visual checks to workflows that already use Jev for structured decisions.

The results are an individual test, not an independent benchmark or evidence of reliable performance across different sites and image types. The strongest reported figures cover just 10 product images. TypeSafe itself warns that the features are not yet ready for production use, so teams considering them would need to test their own examples and decide how to handle uncertain scores or incorrect selections.

Jev’s New Opt-In Image Tools

TypeSafe described Jev 1.14 as a broad set of small improvements and said multilingual support was expected to gain the most. It made the version available as jev-preview and said jev-latest would move to it one week after release, around October 21, 2026. The company also said it would provide 60 days’ deprecation notice for Jev 1.13.0 and 30 days for Jev 1.14.0, with a formal policy still to come.

For image requests, TypeSafe’s format uses a reference in the request and an assets map containing the image. Supported formats are PNG and JPEG, with file and request size limits; WebP and HEIC images must be converted first. TypeSafe says images are billed as ordinary input tokens at Jev’s standard rate, and recommends multipart uploads because base64 encoding increases file size. These implementation details matter to users testing the feature, but do not establish how well it will perform in other settings.

““The results were more honest about my images than I expected.””

— Thorsten Meyer

News-Image Results Are Missing

The supplied source account stops midway through the section on news images, after the heading “News images: Jev was right, my images weren’t.” It gives no scores, examples or conclusion for those 15 images, and does not provide the results for the five essays. It also says both Jev versions were tested but reports only the product-image comparison in the available text. Those omissions prevent a full account of the test or a fair comparison between versions.

The product results come from one site’s images and headlines, with no details in the supplied material about repeat runs, error bars or an external review. The account also does not establish whether a confidence threshold would work reliably beyond the single product-identification miss described. TypeSafe’s warning about production use remains relevant, and the broader accuracy of image judgments and bounding boxes is not established here.

Jev 1.14’s Scheduled Rollout

TypeSafe said jev-preview was available at release and planned to move jev-latest to Jev 1.14 around October 21, 2026. The company’s stated deprecation notice periods are 60 days for Jev 1.13.0 and 30 days for Jev 1.14.0; a formal deprecation policy was still pending in Meyer’s report.

For users, the next practical step is to evaluate the opt-in image and bounding-box features on their own material before relying on them. The full outcome of Meyer’s test cannot be determined from the supplied account: its news-image section is incomplete, and no later test or TypeSafe production-readiness update is included.

Key Questions

What did TypeSafe add to Jev 1.14?

TypeSafe added opt-in image input and a box output type that returns one bounding box and a confidence score. Image input also works on Jev 1.13.0, according to the release information in Meyer’s report.

What did Meyer’s product-image test find?

For 10 product images, Jev 1.14 scored each image’s own headline from 0.65 to 0.89, with a median of 0.76. It picked the right product in nine of 10 cases. The supplied account reports these as results from Meyer’s test, not as a general benchmark.

Did the report show how Jev performed on news images?

No. The source material cuts off just after the news-image section begins. It does not include results for the 15 news images or the five essays in the test.

Is TypeSafe recommending the image feature for production?

No. TypeSafe said the features still had too much “jaggedness” for it to vouch for production use and requires an explicit opt-in header. The company’s caution is separate from Meyer’s favorable product-photo results.

When is Jev 1.14 expected to become jev-latest?

TypeSafe said jev-latest was expected to move to Jev 1.14 about one week after its October 14 release, around October 21, 2026. The release was available as jev-preview in the meantime.

Source: ThorstenMeyerAI.com