🔍 Read the full analysis: Jev’s AI Can See Images Now—So I Tried My Own 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, adding opt-in image input and a bounding-box response type; image input also works with Jev 1.13. In Thorsten Meyer’s 30-image test, Jev 1.14 matched product images to headlines and identified 9 of 10 products, but TypeSafe says the new capabilities are not ready for production use.
TypeSafe released Jev 1.14 on October 14, adding opt-in image input and a new bounding-box response type to its decision model. Thorsten Meyer of ThorstenMeyerAI.com tested the image feature against 30 of his site’s featured images; his early results showed strong product-image matching, but TypeSafe says the capabilities remain too unreliable for production use.
Jev is designed to answer structured questions rather than generate prose: users provide a state and typed questions, and the model returns choices or scores with confidence values. Before this release, the questions described in Meyer’s account were text-based. TypeSafe says image input is now available with Jev 1.14 and Jev 1.13, while the new “box” type is exclusive to 1.14 and returns one bounding box for a specified target, along with confidence.
The image feature is behind the request header “TypeSafe-Danger-Zone-Images: true.” Images can be inserted into the state, question instructions or criteria using references in an assets map. TypeSafe’s documented limits include PNG and JPEG stills, image dimensions from 256 to 1,568 pixels per side, a 16 MiB limit per file and a 32 MiB limit per request. A request fails if an included image is invalid. The company says images are billed as ordinary input tokens, at its listed rate of $0.042 per million input tokens, with no separate image surcharge.
Meyer tested 30 recent images: 10 product roundups, 15 news posts and five essays. He asked whether each image fit its own headline, a randomly selected other headline or an unrelated plant-care headline. For the 10 product images, Jev 1.14 scored the matching headline between 0.65 and 0.89, with a median of 0.76; unrelated headline scores did not exceed 0.10. It selected the correct product in nine of 10 cases. These are results from one small test, not an independent benchmark or a guarantee of performance on other image sets.
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
Where Image Checks Could Help
The release broadens Jev from classifying information presented as text to evaluating visual material alongside structured questions. That could support narrow workflows such as checking whether an image matches a headline, choosing among pictured products or locating a specified object. For publishers and developers, the appeal is a repeatable check that can be run across many assets rather than a manual review of every item.
Meyer’s product-image results suggest the model may be useful for that kind of task, while also showing why confidence values and human checks matter. Jev 1.14 missed one product identification: a photo showing a router beside a cylindrical mesh unit was labeled “Wi-Fi 7 router” with confidence 0.60. In this test, the correct product answers had confidence values of at least 0.77. That pattern is worth testing further, but one set of 10 images does not establish a dependable threshold for other products or use cases.
TypeSafe has placed the feature behind an explicit opt-in rather than changing existing requests by default. That limits accidental use, but also signals that developers should treat the image and bounding-box capabilities as experimental. The company has cautioned that their behavior remains too inconsistent for it to endorse production deployment.
As an affiliate, we earn on qualifying purchases.
A Deliberate Opt-In Release
TypeSafe describes the release as a compromise between developers seeking reliability and users wanting to experiment. Requests without the new headers retain the existing behavior. Image input requires its own danger-zone header; a separate “TypeSafe-Danger-Zone-Long-Context: true” header raises the context limits from 64,000 tokens per request and 32,000 per question to 262,000 for both. TypeSafe advises against the longer setting because reliability can decline on long inputs.
The company’s release post also says Jev 1.13 was multimodal “all along,” meaning image input is not limited to the newly released model. Jev 1.14 itself is described as a set of small, broad improvements, with multilingual support expected to benefit most. The release is available as “jev-preview”; TypeSafe said “jev-latest” would move to it about a week after release, around October 21.
Meyer says he ran roughly 60 requests across Jev 1.13 and 1.14. The supplied account gives detailed results for the product images, including a comparison of product-identification answers: Jev 1.14 got nine of 10 right, while Jev 1.13 got all 10 right but assigned confidence below 0.5 to two. The source text cuts off as it begins describing the results for the 15 news images, so those findings cannot be reported here.
““There is still way too much jaggedness for us to vouch for these for production use cases.””
— TypeSafe, as quoted in its release post
As an affiliate, we earn on qualifying purchases.
How Reliable Are Image Results?
The findings are one publisher’s small, self-selected test and do not establish performance across different image types, languages or real-world workflows. The source provides detailed numbers for the 10 product images but cuts off before reporting the results for the 15 news images and five essays. It also does not provide the full question wording, image set or a third-party evaluation, limiting what readers can independently verify.
TypeSafe has not endorsed the capabilities for production use, and the release account says the company’s internal intelligence estimates may be biased toward Jev. The results do not show how the model performs on difficult crops, low-quality images or images containing several plausible targets. It is also unclear from the supplied material how bounding-box accuracy was evaluated or how often the feature returns a useful location. Developers should not treat the reported confidence scores as calibrated probabilities without further evidence.
As an affiliate, we earn on qualifying purchases.
Release Schedule and Further Testing
TypeSafe said Jev 1.14 is available through “jev-preview” and that “jev-latest” was due to move to it around October 21, one week after release. The company also described deprecation notice periods of 60 days for Jev 1.13.0 and 30 days for Jev 1.14.0, while saying a formal policy is still to come.
For now, developers interested in image input or bounding boxes can test the features with the required opt-in header, while checking results against their own data and task requirements. The next useful evidence would include fuller test results, repeatable evaluations on larger and more varied image sets, and clearer information on the reliability of bounding-box outputs. TypeSafe has not provided a date in the supplied material for when it may consider these features ready for production.
As an affiliate, we earn on qualifying purchases.
Key Questions
What changed in Jev 1.14?
TypeSafe added opt-in image input and a new “box” response type that returns a bounding box and confidence for a specified target. Image input also works with Jev 1.13, according to the company.
What did Meyer’s test find?
Across 10 product images, Jev 1.14 matched each image to its own headline with scores from 0.65 to 0.89 and identified the pictured product correctly in nine cases. The results are from a small, single-site test and are not a general performance benchmark.
Is Jev’s image feature ready for production?
TypeSafe says it is not ready to endorse for production use. Image input requires an explicit danger-zone header, and the company says the new capabilities remain too inconsistent for production use cases.
How much does image input cost?
TypeSafe says there is no separate image surcharge; image data is billed as ordinary input tokens at its listed rate of $0.042 per million input tokens. That is the company’s stated pricing, and actual usage depends on the request.
When will Jev 1.14 become the default latest model?
TypeSafe said “jev-latest” was expected to move to Jev 1.14 about one week after the October 14 release, around October 21, 2026. The release was already available as “jev-preview.”
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
