🔍 Read the full analysis: What Can Jev See In 30 Images I Made? on ThorstenMeyerAI.com
Get hardware and tech essentials 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 added opt-in image input and bounding-box output to Jev, a model designed to answer structured questions. In an author-run test of 30 site images, Jev 1.14 matched all 10 product images to their own headlines at stronger scores than unrelated headlines and identified 9 of 10 products correctly. The test is limited and does not establish production reliability.
TypeSafe’s Jev model now accepts images for structured questions, and a small test by Thorsten Meyer found that the latest version matched product photos to their headlines and identified most pictured products. Meyer tested 30 featured images from ThorstenMeyerAI.com after Jev 1.14’s October 14 release; the results offer an early look at the feature, not evidence that it is ready for production use.
Jev is a decision model that responds to typed questions with outputs such as yes-or-no probabilities, selections from a list, or scores, each paired with a confidence value. Before the release, its questions were limited to text. TypeSafe’s October 14 announcement added image input, available through an explicit request header on both Jev 1.13 and Jev 1.14, and a new bounding-box question type on Jev 1.14 alone.
Meyer evaluated the image feature on the site’s 30 most recent featured images: 10 product roundups, 15 news stories and five essays. For each image, he asked whether it fit its own headline, a randomly selected different headline and an unrelated headline about repotting a monstera. The 10 product images also received a question asking Jev to select which product each showed. Meyer says he made about 60 requests across Jev 1.13 and 1.14.
For the 10 product images, Jev 1.14 assigned each image’s own headline a score from 0.65 to 0.89, with a median of 0.76. The random headline scores reached no higher than 0.18, with a median of 0.06; the plant headline scored no higher than 0.10. Jev 1.14 selected the right product in nine cases. Jev 1.13 got all 10 product selections right, although 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 Product Scores Show
The test suggests that image input could extend Jev’s structured decision-making to tasks such as checking whether a product photo fits a page headline or identifying an item from a fixed list. Those are narrower uses than open-ended image descriptions: the model is being asked to make a defined choice, rather than write a caption or explain a scene.
The results also show why confidence scores may matter in an application. Meyer’s one product-selection error received a confidence of 0.60, while the correct answers in the Jev 1.14 run each had confidence of at least 0.77. That pattern could help a developer flag uncertain responses for human review, but this small set does not establish a generally reliable threshold. A score is the model’s confidence estimate, not a guarantee that an answer is correct.
TypeSafe itself has cautioned against treating the new capabilities as production-ready. The company described the features as experimental and requires users to activate them with a “danger zone” header. That caveat matters for anyone considering automated image checks: a successful test on one site’s images cannot establish performance across different image types, subjects or workflows.
Products Worth Considering
EcoFlow Delta 3 Portable Power Station – 1024 Wh, 1800 W
Anker SOLIX S2000 Solar Powered Generator, 2010 Wh, 1500 W
LG UltraGear 32GX850A-B 32-inch 4K UHD Glossy OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Jev’s Opt-In Image Release
TypeSafe presented Jev 1.14 as a release balancing experimentation with reliability. Image support is available on Jev 1.13 as well as 1.14, which the company said was multimodal already. The new bounding-box output, which returns one box for a described target along with confidence, is limited to 1.14. Both image support and longer context require separate opt-in headers; the default behavior remains unchanged.
Requests can place an image in shared state, question instructions or criteria used for options and scoring. TypeSafe’s documentation, as described by Meyer, allows still images in PNG or JPEG format, with each side between 256 and 1,568 pixels. A file may be up to 16 MiB and a request up to 32 MiB; one invalid image causes the request to fail. The stated price is the usual input-token rate of $0.042 per million input tokens, with no separate image surcharge. Meyer also notes that base64 encoding increases file size and counts toward the request limit.
Jev 1.14 is available under the jev-preview name. TypeSafe said the jev-latest alias would move to it about a week after release, around October 21. The company also raised the optional context limit from 64,000 tokens per request and 32,000 per question to 262,000 for each, while advising against long inputs because reliability can decline. TypeSafe’s internal comparisons were not public benchmark scores; it said they covered open-weight models and could be biased in Jev’s favor.
“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
Limits of the 30-Image Test
Meyer’s evaluation used images from one website and only 10 product photos for product identification. The article excerpt reports the product results but cuts off as it begins discussing the news-image results, so the findings for all 30 images are not available here. It also does not describe a blinded evaluation, an independent review of the answers or a larger test across sites and image categories.
TypeSafe has not vouched for image support in production, and its own release wording identifies reliability issues. The source does not establish how performance changes with image quality, unusual subjects, multiple objects, different prompts or requests containing many images. Nor does the single reported product-selection miss establish that a particular confidence threshold will reliably catch errors elsewhere.
Jev Rollout and Further Testing
TypeSafe said it expected the jev-latest alias to point to Jev 1.14 around October 21, 2026, after a week on jev-preview. It also said users would receive 60 days’ deprecation notice for Jev 1.13.0 and 30 days for Jev 1.14.0, with a formal policy still to come.
For developers testing the image feature, the immediate next step is to evaluate it on the images and questions they intend to use, while accounting for the opt-in status, supported formats and request limits. Meyer said TypeSafe’s internal estimates should be checked against independent evaluations. The company has not announced when, or whether, it will consider the image feature ready for production.
Key Questions
What changed in Jev 1.14?
TypeSafe released Jev 1.14 on October 14, 2026. The release includes image input behind an opt-in header, a new bounding-box question type and other broad improvements. Image input also works on Jev 1.13, according to the company.
How did Jev perform on the 10 product images?
In Thorsten Meyer’s test, Jev 1.14 selected the correct product for nine of the 10 images. Its score for each image’s own headline ranged from 0.65 to 0.89, while scores for random and unrelated headlines were lower in this sample.
Does the test show Jev is ready for production?
No. The evaluation was a small, author-run test on images from one website. TypeSafe said it could not yet vouch for the new capabilities in production use cases and made them opt-in.
What image files does Jev accept?
The source says the feature accepts PNG and JPEG still images, with each side measuring 256 to 1,568 pixels. It lists a 16 MiB limit per file and 32 MiB per request; WebP and HEIC images need conversion first.
When does Jev 1.14 become the latest model?
TypeSafe said Jev 1.14 would become the jev-latest model about one week after release, around October 21, 2026. It was available as jev-preview when the source article was published.
Source: ThorstenMeyerAI.com
