🔍 Read the full analysis: Exploring Jev’s AI Vision Through 30 Images I Made on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer tested TypeSafe’s image-capable Jev model on 30 featured images from his website, asking whether each image fit its headline and, for product photos, which product it showed. He reports perfect separation on 10 product images with Jev 1.14 and correct product identification in nine of 10 cases; the supplied source cuts off before giving the news-image results.
Thorsten Meyer tested TypeSafe’s newly image-capable Jev decision model on 30 featured images from his website, reporting strong results on product photos but an incomplete account of its news-image performance in the available source. The test follows TypeSafe’s October 14 release of Jev 1.14, which added image input and a bounding-box response type behind explicit opt-in headers.
Meyer, writing on ThorstenMeyerAI.com on October 15, said he tested the 30 most recent featured images on the site: 10 product roundups, 15 news posts and five essays. For each image, he asked whether it fit its own headline, whether it fit a randomly selected other headline, and whether it fit an unrelated headline about repotting a monstera plant. The product images also received a multiple-choice question asking which of 10 products each showed.
In Meyer’s account, Jev 1.14 separated the product images from the comparison headlines: own-headline scores ranged 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, while the plant headline scored no higher than 0.10. These figures describe this small test set, not a general benchmark.
Jev 1.14 identified the product correctly in nine of 10 images, Meyer said. The exception showed a mesh Wi-Fi system with both a router and a cylindrical mesh unit; the model selected “Wi-Fi 7 router” with 0.60 confidence. Meyer said all correct product selections had confidence of at least 0.77. Jev 1.13 identified all 10 products, but two answers had confidence below 0.5.
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
What the Image Test Can Show
The test offers an early, practical example of how a model designed to make structured decisions might be used with visual material. A publisher could ask whether an image appears relevant to a headline or use a model to select among product labels. Meyer’s results suggest those checks may be worth evaluating, especially when a workflow can route uncertain answers to a person rather than treating every output as reliable.
But the findings are one creator’s test of 30 site images, not an independent assessment across publishers, image styles or subject areas. The difference between Jev 1.13 and 1.14 on product identification also shows why confidence and version should be tracked alongside the answer. TypeSafe itself warned that the new features are not ready for production use without further evaluation.
As an affiliate, we earn on qualifying purchases.
Jev’s New Opt-In Image Features
Before this release, Jev accepted typed questions about text. The model returns structured decisions, such as a yes-or-no probability, a choice from a list or a score, each with a confidence value. TypeSafe’s October 14 post said image support also works with Jev 1.13.0, describing that version as multimodal even though image use had not been announced earlier.
Image requests require the TypeSafe-Danger-Zone-Images: true header. The release also adds a “box” response type in Jev 1.14, which returns a bounding box and confidence for a specified target. A separate opt-in header raises the context limit from 64,000 to 262,000 tokens per request, though TypeSafe advises against the larger limit because reliability can drop on long inputs.
TypeSafe said the new capabilities retain the existing production behavior unless users opt in, and warned that there is “still way too much jaggedness” to recommend them for production use. Image inputs are restricted to PNG and JPEG stills between 256 and 1,568 pixels per side, with limits of 16 MiB per image and 32 MiB per request, according to Meyer’s account of the 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
The Missing News-Image Results
The available source ends as Meyer begins describing the 15 news-post images, after the words “News images: Jev was right, my images weren’t a dif”. It does not provide the rest of that section, so the news-image scores, examples and Meyer’s conclusion about them cannot be reported here. The source also does not give the results for the five essays separately or a complete table of all requests.
The test was not described as a controlled or independent benchmark. Its results apply to the images and questions Meyer selected, and the source does not establish how Jev would perform on other websites or image categories. TypeSafe has also cautioned against relying on the new capabilities in production.
Jev’s Rollout and Further Testing
TypeSafe said jev-preview was available with Jev 1.14 at release. According to Meyer, the jev-latest alias was due to move to that version about one week later, around October 21, 2026. TypeSafe also said it would give 60 days’ deprecation notice for version 1.13.0 and 30 days for version 1.14.0, with a formal policy still to come.
For readers assessing the image feature, the next useful evidence would include the missing news-image findings and tests on larger, more varied image sets. TypeSafe advises users to run their own evaluations, and its release warning means the opt-in capabilities should be treated as experimental rather than established production tools.
Key Questions
What did Thorsten Meyer test?
He tested whether 30 featured images from his website fit their headlines, a randomly selected alternative headline and an unrelated plant-care headline. He also asked Jev to identify the products shown in 10 product images.
How did Jev 1.14 perform on the product images?
According to Meyer, Jev 1.14 separated each of the 10 product images from the comparison headlines and identified the product correctly in nine of 10 cases. This was a small, site-specific test, not a general performance benchmark.
Did Jev 1.13 also support images?
TypeSafe said image support works on jev-1.13.0 as well as Jev 1.14. Meyer reported that Jev 1.13 identified all 10 products in his test, though two answers had confidence below 0.5.
Can developers use Jev’s image features in production?
TypeSafe has not vouched for the new capabilities in production use and requires users to opt in with a danger-zone header. The company advised caution because reliability issues remain.
What were the results for the news images?
The supplied source cuts off as Meyer begins that section, so its news-image results are not available in the material provided. No scores or conclusions for those 15 images can be confirmed from it.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
