I Put My Own Pictures To Work In A Jev Vision Test

🔍 Read the full analysis: I Put My Own Pictures To Work In A Jev Vision Test on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• 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 analysis and a new bounding-box response type to its Jev decision model on October 14, 2026. The next day, Thorsten Meyer tested the release on 30 featured images from ThorstenMeyerAI.com, reporting strong results for product photos while cautioning that TypeSafe itself does not yet recommend the new capabilities for production use.

Jev is designed to answer structured questions rather than generate prose: users provide a state and typed questions, and the model returns responses such as a yes-or-no probability, a choice from a list, or a score, along with confidence. Before this release, the questions in the account had to concern text. TypeSafe says image support also works with jev-1.13.0, while the new box type is limited to Jev 1.14.

Image input is enabled with the request header TypeSafe-Danger-Zone-Images: true. Images can be placed in the request state, a question’s instructions, or criteria used to define choices and score levels. The release also adds a separate opt-in for longer context, raising the stated request and question limits from 64,000 and 32,000 tokens to 262,000 tokens. TypeSafe cautions that reliability declines on long inputs and advises against relying on the expanded limit without testing.

Meyer used the 30 most recent site images: 10 from product roundups, 15 from news posts and five from essays. For each, he asked whether it fit its own headline, a randomly selected other headline and an unrelated plant-care headline. The 10 product images also received a multiple-choice identification question. Meyer says the test involved about 60 requests across Jev 1.14 and 1.13, but the supplied account cuts off while beginning to describe the news-image results.

For the product images, Meyer reports that Jev 1.14 assigned each image’s own headline a score from 0.65 to 0.89, with a median of 0.76. Random alternative headlines had a median score of 0.06 and did not exceed 0.18; the unrelated plant headline stayed at or below 0.10. Jev 1.14 identified the product correctly in nine of 10 cases. The one miss was a mesh Wi-Fi image that showed both a router and a cylindrical mesh unit; the model selected “Wi-Fi 7 router” with 0.60 confidence.

A Test of Image Matching in Practice

The test offers an early, small-scale example of how a decision model might help with a narrow editorial task: checking whether a featured image relates to a headline. Meyer says this kind of review is often not done by hand, and his product-photo results suggest that structured image questions can separate relevant examples from mismatches in this particular set.

The results are not a general measure of Jev’s image ability. The sample came from one website, the questions were written by the author, and there is no independent evaluation described in the source. The figures are observations from that test, not evidence that the model will perform similarly on other image collections. TypeSafe’s own warning that the capabilities remain too unreliable for production is a further reason to treat the results as preliminary.

For users considering image-based automation, the release’s opt-in design makes the boundary clear: existing requests do not automatically begin using the new features. Teams can test the functions separately, but should check outputs and failure cases before relying on them. The reported confidence scores may help flag uncertain answers in Meyer’s sample, but that does not establish a dependable threshold for other tasks.

How TypeSafe Introduced Image Input

TypeSafe presented the release as a balance between reliability for builders and experimentation for users. Its announcement says the new capabilities are behind explicit request headers because the company sees too much “jaggedness” to endorse them for production. The term “danger zone” is part of the header naming; it does not mean the features are enabled by default.

Images are referenced through an assets map in the request, and a reference string is replaced wherever it appears in the state or instructions. TypeSafe advises using a fresh UUID as the reference because matching is plain text. Placing an image in the state allows it to be used across questions, while putting it into individual instructions can send it repeatedly and affect input use.

The current limits described in the source are PNG and JPEG still images between 256 and 1,568 pixels per side, with a 16 MiB maximum per file and 32 MiB per request. WebP and HEIC files must be converted; a single invalid image can fail the whole request. TypeSafe says images are billed as ordinary input tokens at Jev’s stated rate of $0.042 per million input tokens, with no separate image surcharge. The source also says Jev 1.14 is available as jev-preview, with jev-latest expected to move to it about a week after release.

“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 as quoted by Thorsten Meyer

What the Image Test Leaves Open

The supplied source text ends just as Meyer starts the section on news-image results. It says the news posts were a “dif” before cutting off, so it does not provide the findings for the 15 news images, the five essays, or the full results for Jev 1.13 across the headline-matching questions. Those outcomes cannot be inferred from the product-photo figures.

The account also does not give image-by-image scoring data, an independently checked answer key, or details about whether the test was repeated. It describes the model’s performance on one site’s selected images, not a representative benchmark. TypeSafe’s warning about production reliability remains in force, and the source does not establish how the new box output performs on real-world detection tasks.

It is also unclear from the supplied material how TypeSafe will formalize its model deprecation policy. Meyer reports that the company gave 60 days’ notice for Jev 1.13.0 and 30 days for 1.14.0, with a formal policy still to come.

Further Testing and Model Rollout

TypeSafe said Jev 1.14 was available as jev-preview and planned to move the jev-latest alias to it around October 21, 2026, roughly one week after the October 14 release. Users evaluating image input or bounding-box output will need to opt in explicitly and test the capabilities against their own tasks.

The next useful information would be the complete results from Meyer’s news-image and essay groups, along with further evaluations that include more images and independent checks. TypeSafe also said a formal deprecation policy was still to come. Until then, the company’s stated reliability caveat means the release is best understood as an available feature for testing, not a production endorsement.

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 score for a described target. The box type is available only in Jev 1.14, according to the source.

What did Meyer’s image test find?

For 10 product images, Meyer reports that Jev 1.14 matched each image to its own headline with scores from 0.65 to 0.89 and identified the product correctly in nine of 10 cases. The provided source cuts off before giving the results for the news images and essays.

Does TypeSafe recommend using image support in production?

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

Which image formats and sizes are supported?

The source lists PNG and JPEG still images between 256 and 1,568 pixels per side, with a limit of 16 MiB per image and 32 MiB per request. It says WebP and HEIC files need to be converted first.

When was Jev 1.14 expected to become jev-latest?

TypeSafe’s stated plan was to move the jev-latest alias to Jev 1.14 about one week after the October 14, 2026 release, around October 21. The source describes that as a planned rollout.

Source: ThorstenMeyerAI.com