🔍 Read the full analysis: Jev Meets My Image Library: A 30-Picture Trial on ThorstenMeyerAI.com
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
TL;DR
TypeSafe released Jev 1.14 on October 14, 2026, adding opt-in image inputs and a bounding-box answer type while keeping the features outside default production behavior. In a test of 30 images from ThorstenMeyerAI.com, the author reported strong product-image matching but mixed results across news images; the supplied account cuts off before giving the full news-image findings.
Jev is designed to return structured decisions rather than write prose: users provide a state and typed questions, and the model responds with choices, scores or probabilities and confidence. Before this release, those questions were limited to text, according to Thorsten Meyer’s report on ThorstenMeyerAI.com. TypeSafe says image support also works with Jev 1.13.0, describing that version as having been “multimodal all along.”
For image requests, a developer places an image in an assets map and inserts its reference wherever the image should be used—in the state, question instructions or scoring criteria. The release also introduces a box type in Jev 1.14, which returns one bounding box for a described target and a confidence score. Image input accepts still PNG and JPEG files within stated size and dimension limits; unsupported formats such as WebP and HEIC need conversion.
Meyer tested the 30 most recent featured images on his site: 10 product roundups, 15 news posts and five essays. He asked three yes-or-no questions about each image, including whether it matched its own headline, a randomly selected other headline and an unrelated plant-care headline. Product images also received a question asking Jev to identify one of 10 products. The report says Jev 1.14 correctly identified nine of the 10 products and separated matching from mismatched headlines in that product subset.
Jev 1.14 · images + box
Jev can see now. Here is what 30 of my images showed.
I asked about every featured image three times: does it fit its own headline, a random other post’s headline, an off-topic headline? Both Jev versions, about 60 requests.
Does this image fit the headline?
Median “yes” probability on Jev 1.14, by kind of post
- Product roundups10 images0.760.061.13: 0.79 / 0.05
- News posts15 images0.330.161.13: 0.51 / 0.25
- My essays5 images0.510.161.13: 0.62 / 0.33
Off-topic headline (“How to Repot a Monstera Plant”): never above 0.10 on either version, 60 of 60 rejected.
Why news is weak: almost all news images show the same scene, people at desks looking at monitors. Jev could not tell them apart, and it was right not to.
1.14 is stricter than 1.13
Same 30 images, same questions
| Result | 1.14 | 1.13 |
|---|---|---|
| Fits own headline (≥ 0.5) | 15/30 | 23/30 |
| Fits a wrong headline (≥ 0.5) | 1/30 | 4/30 |
| Own beats random other (AUC) | 0.91 | 0.89 |
| Names the product (10 options) | 9/10 | 10/10 |
1.14’s one product miss: mesh system read as “Wi-Fi 7 router”, confidence 0.60. Re-tune thresholds before jev-latest moves.
What shipped on Oct 14
- headerImagesin state, instructions or criteria · 1.13 + 1.14 ·
TypeSafe-Danger-Zone-Images - 1.14 onlyBox typeone bounding box + confidence. 6 of 6 clear targets boxed (0.66–0.90); an absent bicycle still got a box (0.03)
- headerLong context64k / 32k → 262k tokens, slower and less accurate
- opt-inHelp find weaknessesqueries used for testing, not training
now: jev-preview = 1.14~Oct 21: jev-latest → 1.14notice: 60 days (1.13), 30 days (1.14)
Source: own test on October 15, 2026. 30 featured images from ThorstenMeyerAI.com (10 product roundups, 15 news posts, 5 essays), JPEG at up to 1,568 px, jev-1.14.0 and jev-1.13.0 with the image header. Bars show medians. AUC = probability that an image’s own headline scores higher than a random other headline. Features, prices and limits: TypeSafe blog post “Jev 1.14” and docs.typesafe.ai, as of October 15, 2026.
Powered by Thorsten Meyer AI
Image Decisions Require Careful Testing
The release expands Jev from text-based classification to image-based decisions, which could support tasks such as checking whether a picture matches a headline or locating an object in an image. Those are practical uses for a model built to produce structured answers, rather than open-ended descriptions.
But the trial is a small, publisher-run evaluation using images from one website, not an independent benchmark. The reported product results suggest promise for that narrow task; they do not establish performance across image types, industries or production settings. TypeSafe itself says the new capabilities remain too jagged for it to recommend production use, and keeps them behind explicit opt-in headers.
That caution matters to developers deciding whether to rely on model outputs. A confidence score may help flag uncertain answers, as Meyer says it would have caught the one product-identification miss in his sample. It does not by itself show that a confidence threshold will work reliably on other datasets.
As an affiliate, we earn on qualifying purchases.
How TypeSafe Is Limiting Release Risk
TypeSafe presents Jev 1.14 as a release that preserves existing default behavior while allowing experimentation through request headers. Image support requires TypeSafe-Danger-Zone-Images: true; longer context requires a separate opt-in. The company labels both features experimental and says it will not vouch for them in production use cases yet.
The standard request limits remain 64,000 tokens overall and 32,000 per question. An opt-in header raises those limits to 262,000, though TypeSafe advises against longer inputs because reliability can decline. The release also includes broad model improvements, with multilingual support expected by TypeSafe to benefit most. Its `jev-preview` model is available now, and `jev-latest` is scheduled to move to it about a week after release.
TypeSafe shared internal intelligence estimates rather than public benchmark scores, Meyer wrote, and cautioned that the comparisons against open-weight models may favor Jev. He recommends users run their own evaluations. The trial described here is one such evaluation, but its partial published account limits what can be concluded.
““was multimodal all along””
— TypeSafe, describing Jev 1.13.0
As an affiliate, we earn on qualifying purchases.
The Trial’s Missing Results Matter
The source account supplied for this article ends in the middle of Meyer’s discussion of the 15 news images, after the line “The news posts were a dif”. It does not include the results for news images or the five essays, a final comparison between Jev 1.14 and 1.13 across the full trial, or any later conclusions. Those findings cannot be inferred from the product-image results.
The test used images from one site and a relatively small sample, and the source does not provide an independent evaluation or details sufficient to assess how representative the sample was. It is also unclear from the supplied material whether the reported scores were measured under a fixed evaluation procedure beyond the questions described.
TypeSafe’s own production-readiness warning remains in place. The company has not said when the image and bounding-box features will leave experimental status, and the report provides no evidence that the trial’s results generalize to other users’ image libraries.
As an affiliate, we earn on qualifying purchases.
Watch Opt-In Testing and Model Rollout
Developers can test image input and bounding boxes through the opt-in headers, while keeping in mind TypeSafe’s warning about reliability. The company recommends users evaluate Jev on their own tasks rather than relying on its internal comparisons or a single small trial.
TypeSafe planned to move `jev-latest` to Jev 1.14 around October 21, one week after the release. Its stated deprecation policy gives 60 days’ notice for Jev 1.13.0 and 30 days for Jev 1.14.0, with a formal policy still to come. Further details on the image features’ reliability and production status have not been provided in the source material.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did TypeSafe add in Jev 1.14?
Jev 1.14 adds image inputs and a bounding-box output type. Both capabilities require explicit opt-in; TypeSafe says they are not yet suitable for production use cases it can vouch for.
What did the 30-image trial find?
In the reported product-image subset, Jev 1.14 matched images to their own headlines better than to unrelated headlines and identified nine of 10 products correctly. The supplied source ends before giving the findings for news images and essays.
Does Jev image support work with Jev 1.13?
According to TypeSafe, image support also works with jev-1.13.0, which the company described as having been multimodal all along. The new bounding-box type is available only in Jev 1.14.
Can developers use the features in production?
They can access the features by opting in, but TypeSafe says there is too much reliability “jaggedness” for it to vouch for production use. Developers should test performance on their own data and use cases.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
