🔍 Read the full analysis: How Jev Described 30 Images I Shared on ThorstenMeyerAI.com
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
TL;DR
Thorsten Meyer tested TypeSafe’s Jev decision model on 30 images from his site after image input arrived in Jev 1.14. He reported strong results for product images, including nine correct product identifications out of 10, but the supplied source cuts off before giving the results for news images and essays.
Thorsten Meyer tested TypeSafe’s Jev decision model on 30 featured images from ThorstenMeyerAI.com after the company added image input in Jev 1.14, reporting strong headline-matching results for product photos. In the product-image portion, Jev 1.14 selected the correct product for nine of 10 images; the source account provided here ends before detailing the news-image and essay results.
TypeSafe published Jev 1.14 on October 14, 2026, adding support for image inputs and a new bounding-box response type. The company also made the new capabilities opt-in through request headers, saying they remain too inconsistent for it to recommend them in production. Image support also works with Jev 1.13, according to TypeSafe’s release notes as described by Meyer.
Meyer tested 30 recent site images: 10 product roundups, 15 news posts and five essays. For each, he asked whether the image fit its own headline, a randomly chosen other headline, and an unrelated headline about repotting a monstera plant. He also asked Jev to identify which of 10 products appeared in each product image. The test used Jev 1.14 and 1.13, with about 60 requests in total.
In the product-image results reported in the available source, Jev 1.14 gave each image’s own headline a fit score from 0.65 to 0.89, with a median of 0.76. Random headline scores reached no more than 0.18, with a median of 0.06, and the plant headline scored no higher than 0.10. Jev 1.14 identified nine of the 10 products correctly. Meyer said the incorrect answer involved an image showing both a router and a mesh unit; Jev selected “Wi-Fi 7 router” with confidence 0.60.
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 Shows
The test offers an early, practical example of how a structured decision model can assess images without generating a free-form description. Meyer’s task was narrow: judge whether a featured image fits a headline, then classify product photos. That makes the reported results relevant to publishers considering automated checks for image selection or labeling.
But the findings are one person’s test on 30 images from one website, not a broad evaluation of image understanding. The source reports detailed numerical results only for product images before cutting off during its discussion of news images. It does not establish how well Jev would perform on other publishers’ libraries, different image types or a larger, independently selected sample.
TypeSafe’s own caution also matters. The company placed image support behind an explicit opt-in and said it could not yet vouch for the feature in production. Meyer’s results are an account of his experiment, not a guarantee of accuracy. Publishers would need to test the model on their own material and decide how to handle uncertain outputs, including whether a human should review them.
As an affiliate, we earn on qualifying purchases.
How Jev Added Image Input
Before the release, Jev’s questions were limited to text, according to Meyer. The model takes a state and typed questions, then returns outputs such as a yes-or-no probability, a choice from a list or a score, each with a confidence value. TypeSafe describes this format as a decision model rather than a general-purpose writing system.
For image requests, TypeSafe uses an assets map: an image is attached to a reference string, and that string is placed where the image should be used in the state or question. The release also added a box response type in Jev 1.14, which returns one bounding box for a described target along with confidence. Image input is enabled with the header TypeSafe-Danger-Zone-Images: true.
The release included other changes, too. A separate opt-in header raises the context limit from 64,000 tokens per request and 32,000 per question to 262,000, although TypeSafe advises against longer inputs because reliability can decline. The new model is available as jev-preview; TypeSafe said jev-latest would move to it about a week after release, around October 21.
“There is “still way too much jaggedness for us to vouch for these for production use cases.””
— TypeSafe, as quoted in Meyer’s account of the release
product image identification tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Results Beyond Product Photos
The source material supplied for this article stops mid-sentence as Meyer begins the section on news-image results. It does not provide the scores or conclusions for the 15 news images, nor the five essays, so no overall result for all 30 images can be stated. The account also does not give the full prompts, image-by-image outputs or an independent review of the test.
It is also unclear how results might change with a different sample or model settings. Meyer reported that Jev 1.13 identified all 10 products correctly but assigned confidence below 0.5 to two of them; the available text gives no comparable full set of headline-fit scores for that version. TypeSafe’s warning about production use remains in place.
As an affiliate, we earn on qualifying purchases.
Release Rollout and Further Testing
TypeSafe’s announced rollout puts jev-1.14.0 on jev-preview first, with jev-latest expected to move to the new version about one week after the October 14 release. The company said it would provide 60 days’ deprecation notice for Jev 1.13.0 and 30 days for 1.14.0, with a formal deprecation policy still to come.
For readers evaluating the image feature, the next useful step is to compare its outputs against a representative image set and retain human review where mistakes matter. Meyer said TypeSafe’s internal estimates were based on comparisons with open-weight models and might favor Jev, and advised users to run their own evaluations. More complete results from his 30-image test would be needed to judge performance across the news and essay categories.
image captioning and tagging software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Jev add in version 1.14?
Jev 1.14 added image input and a bounding-box response type. TypeSafe made the features opt-in, citing reliability concerns for production use.
How many images did Meyer test?
Meyer said he tested 30 featured images from his site: 10 product roundups, 15 news posts and five essays, using Jev 1.14 and 1.13.
How did Jev 1.14 perform on product images?
In the reported product test, Jev 1.14 identified the right product in nine of 10 images. Its own-headline fit scores ranged from 0.65 to 0.89, according to Meyer.
Are the results available for all 30 images?
No. The supplied source ends as Meyer begins discussing the news-image results. It does not include results for the 15 news images or five essays.
Is TypeSafe recommending Jev image support for production?
No. TypeSafe said the feature still had too much inconsistency for production use and placed it behind an explicit opt-in header.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
