🔍 Read the full analysis: Does Jev Get What’s In My Images? I Tested 30 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
TypeSafe released Jev 1.14 on October 14, 2026, adding opt-in image input and a bounding-box answer type. Thorsten Meyer tested the image feature on 30 images from his site; product-image results were strong, but the supplied source cuts off before reporting the news-image and overall findings.
TypeSafe’s Jev 1.14, released October 14, can accept images in decision-model requests, a change Thorsten Meyer tested on 30 featured images from his website. In the reported product-image results, Jev 1.14 matched images to their own headlines more strongly than to unrelated headlines, but TypeSafe says the new capabilities remain experimental and should be explicitly enabled.
Meyer compared Jev 1.14 with Jev 1.13 using three yes-or-no questions for each image: whether it fit its own post headline, a randomly selected different headline, and an unrelated headline about repotting a monstera plant. The sample comprised 10 product roundups, 15 news posts and five essays. The report says the test involved about 60 requests overall.
For the 10 product images, Jev 1.14 gave each image’s own headline a score of 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; the plant headline stayed at or below 0.10. On a separate question asking it to identify which of 10 products appeared in each image, Jev 1.14 was right on nine of 10.
The one product identification miss involved a mesh Wi-Fi image showing an antenna-equipped router beside a cylindrical mesh unit. Jev named it a “Wi-Fi 7 router” with 0.60 confidence. Meyer says all correct answers in that test had confidence of at least 0.77. Jev 1.13 identified all 10 products correctly, though two answers had confidence below 0.5. These are results from a small, single-site test, not a general benchmark.
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 Checks Move Into Jev
The release changes Jev from a decision model that, according to Meyer’s account of its earlier limits, handled questions about text into one that can use images as input. That could support narrow checks such as whether a picture matches a headline, or whether a particular object appears in an image. The reported product results suggest those tasks may be useful to test, but the sample is too small to establish how the model performs across other image types, subjects or workflows.
TypeSafe has put the feature behind an explicit danger-zone request header and cautions that its output is not reliable enough for production use cases. That warning matters for anyone considering automated image review: a confident answer is not proof that a classification is correct, and the test itself includes a product misidentification. The results support further evaluation, not reliance without checks.
Popular right now
Bose QuietComfort Wireless Noise Cancelling Headphones, Black
Ninja Foodi DZ201 8-Quart DualZone 6-in-1 Air Fryer
Shark HydroVac MessMaster Wet/Dry Vacuum & Mop Combo (AW261)
As an affiliate, we earn on qualifying purchases.
What Jev 1.14 Adds
Jev is described in Meyer’s report as a model for structured decisions rather than open-ended writing. Users supply a state and typed questions; Jev can return a probability for a yes-or-no question, select from options, or score against a scale, with confidence information. Before the October 14 update, Meyer says these questions had to concern text.
TypeSafe’s release adds image input through an assets map: a request defines an image once under a reference, then uses that reference in the state, instructions or criteria. Images are supported on Jev 1.13.0 and 1.14.0 when the image feature is enabled. A separate new box answer type, available only in 1.14, returns one bounding box and a confidence for a described target.
The image feature accepts PNG and JPEG still images between 256 and 1,568 pixels per side, with a 16 MiB limit per file and 32 MiB per request, according to the source report. It does not automatically resize images, and a request fails if one image is invalid. TypeSafe also retains a default context limit of 64,000 tokens per request and 32,000 per question; a separate opt-in header raises those limits, with the company warning that reliability falls on long inputs.
““The mesh Wi-Fi photo shows a router with antennas next to a cylindrical mesh unit, and Jev answered ‘Wi-Fi 7 router’ with a confidence of 0.60.””
— Thorsten Meyer
News Image Results Are Missing
The supplied source ends just as Meyer begins discussing the 15 news images, after the line “News images: Jev was right, my images weren’t a dif”. It does not provide the results for that group, the five essays, the full comparison between model versions on the headline-fit questions, or a complete conclusion. Those findings cannot be inferred from the product-image results.
The test uses 30 images from one website and is not a representative benchmark. The source also does not provide independent validation, details about repeated runs or evidence that the scores predict performance in other settings. TypeSafe’s own caution is that the experimental capabilities are not ready for production use, and its internal estimates are not presented as public benchmark results.
More Testing Before Production
TypeSafe made Jev 1.14 available as jev-preview after the October 14 release. According to Meyer’s report, the jev-latest alias was due to move to 1.14 about one week later, around October 21. TypeSafe also gave 60 days’ deprecation notice for Jev 1.13.0 and 30 days for 1.14.0, while saying a formal deprecation policy was still to come.
For image use, the next practical step is evaluation against a wider and independently checked set of examples, with the model’s confidence compared against actual errors. TypeSafe’s stated production caution remains in force; the supplied material does not report a later change to that guidance or provide the missing results from Meyer’s news-image test.
Key Questions
What changed in Jev 1.14?
TypeSafe added opt-in image input, a new bounding-box answer type in Jev 1.14, and an option to raise context limits. The company cautions that the new capabilities are not yet suitable for production use cases.
How did Jev 1.14 perform on the product images?
In Meyer’s test of 10 product images, Jev 1.14 selected the correct product for nine. It also scored each image’s own headline above randomly chosen alternatives in the reported results. The test was limited to images from one website.
Did Jev 1.13 do better than Jev 1.14?
Jev 1.13 identified all 10 products correctly, compared with nine correct answers for 1.14. But two of the 1.13 answers had confidence below 0.5. The supplied report does not include the full results for the other test questions, so it does not establish which version performed better overall.
Are the image-test results conclusive?
No. They cover a 30-image sample from one site, and the supplied source cuts off before reporting the results for news images and essays. TypeSafe also says the image capabilities are too inconsistent for it to vouch for them in production.
When was Jev 1.14 expected to reach jev-latest?
Meyer’s October 15 report said TypeSafe planned to move jev-latest to Jev 1.14 about a week after the October 14 release, around October 21. The supplied material does not confirm whether that later update occurred.
Source: ThorstenMeyerAI.com
