🔍 Read the full analysis: I Used 30 Personal Images To Explore Jev’s New Vision 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 released Jev 1.14 on October 14, 2026, adding opt-in image input and a bounding-box response type. Thorsten Meyer tested the image features on 30 images from his site and reported strong separation on product photos, while warning that the test was small and TypeSafe does not recommend the new capabilities for production yet.
TypeSafe released Jev 1.14 on October 14, adding image input and a bounding-box output option to its structured decision model. In a test published the following day, Thorsten Meyer ran the image feature against 30 featured images from his website and found that it distinguished product photos from mismatched headlines, while TypeSafe cautioned that the new capabilities are not ready for production use.
Jev is designed to answer structured questions rather than generate ordinary prose. Before this release, its questions were limited to text. TypeSafe’s update allows an image to be supplied in the request’s state, a question’s instructions, or the criteria used to define answer options. The feature requires the explicit header TypeSafe-Danger-Zone-Images: true, and works with both Jev 1.13.0 and 1.14.0.
Jev 1.14 also adds a box response type that returns one bounding box for a described target, along with a confidence score. That feature is limited to Jev 1.14. The release includes a separate opt-in for longer context, increasing request and question limits from 64,000 and 32,000 tokens to 262,000 each. TypeSafe says reliability can decline with long inputs and advises against using the larger limits without testing.
Meyer tested the image feature on 10 product roundups, 15 news posts and five essays. For each image, he asked whether it fit its own headline, a randomly selected other headline and an unrelated plant-care headline. The 10 product images also received a question asking which product each showed. He reports about 60 requests across Jev 1.13.0 and 1.14.0. The supplied account gives detailed product-image results but cuts off before reporting the full results for news images.
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 for Repetitive Work
The release gives developers a way to use Jev for narrow visual classification tasks, such as checking whether a featured image matches a headline or identifying an object in a picture. That could make it easier to apply the model to recurring review work that is costly or inconsistent to do manually. Meyer’s test offers an early example, not evidence that the feature will perform similarly across other sites, image types or higher-stakes tasks.
The results also show why confidence scores and human review matter. On the product-identification question, Jev 1.14 named the correct product in nine of 10 images. The incorrect answer had a confidence score of 0.60, while correct answers scored at least 0.77 in this small sample. That pattern suggests a threshold might have flagged the error in this test, but it does not establish a reliable cutoff for other uses.
TypeSafe’s opt-in design makes the limitations part of the release itself. Image handling and longer context are behind warning-style headers, and the company says there is too much inconsistency to endorse the new capabilities for production. Teams evaluating them need to test their own data and account for failures rather than treating the reported results as a guarantee.
Products Worth Considering
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and 12MP Center Stage Camera (2026, Silver)
Trezor Safe 5 Crypto Hardware Wallet with Color Touchscreen
Ledger Stax Crypto Wallet with Touchscreen
As an affiliate, we earn on qualifying purchases.
How TypeSafe Is Limiting Rollout
TypeSafe describes the release as a balance between developers who prioritize reliability and users who want to experiment. Existing production behavior remains unchanged unless a request includes the relevant opt-in header. The company says Jev 1.13.0 was already multimodal, although image support had not previously been documented as a usable capability. Jev 1.14 makes the feature explicit and adds the new bounding-box type.
Images are referenced through an assets map in a request, with a text reference marking where each image should be inserted. The reference is matched as plain text, so TypeSafe recommends using a fresh UUID rather than a short string that might appear elsewhere. Supported still images are PNG and JPEG, with sides between 256 and 1,568 pixels. Each file is limited to 16 MiB and each request to 32 MiB; one invalid image causes the entire request to fail.
TypeSafe says there is no separate image surcharge: image data is billed as ordinary input tokens at the model’s standard rate of $0.042 per million input tokens, while output tokens are free. Base64 encoding increases file size by about a third, and that extra size counts toward request limits. TypeSafe recommends multipart uploads, which its Python and JavaScript software development kits support.
“There is still way too much jaggedness for us to vouch for these for production use cases.”
— TypeSafe, in its release announcement
Limits of the 30-Image Test
The test covers a small collection from one website, and the source material does not provide the full results for its 15 news images or five essays. It also does not report a formal benchmark, independent replication, or testing on varied image sources. Meyer’s product-image findings should be read as one site’s preliminary evaluation, not a general measure of Jev’s visual accuracy.
TypeSafe has not vouched for image support or longer context in production use. The release account does not establish how performance varies across image quality, complex scenes, languages or different bounding-box targets. It also remains unclear what reliability threshold TypeSafe would consider sufficient, or when it may remove the danger-zone requirements.
Jev 1.14 Rollout Schedule
Jev 1.14.0 is available through the jev-preview model name. TypeSafe said the jev-latest alias would move to the new release one week after launch, around October 21, 2026. Developers considering the image feature can test it with the opt-in header, while keeping in mind the file limits and the company’s warning about production reliability.
TypeSafe said Jev 1.14 is a small, broad update and expects multilingual support to gain the most. The company also described its internal intelligence estimates as comparisons only with open-weight models and warned that those estimates may favor Jev. It recommends that users run their own evaluations. TypeSafe has given 60 days’ deprecation notice for Jev 1.13.0 and 30 days for Jev 1.14.0, with a formal deprecation policy still to come.
Key Questions
What did TypeSafe add to Jev 1.14?
Jev 1.14 adds opt-in image input and a box response type that returns one bounding box and a confidence score. Image input also works on Jev 1.13.0, according to TypeSafe.
What did the 30-image test find?
Meyer reports that Jev 1.14 separated product images from mismatched headlines in his test and identified the correct product in nine of 10 cases. The reported results come from one website and a small sample.
Is TypeSafe recommending the image feature for production?
No. TypeSafe says there is too much inconsistency to vouch for the new capabilities in production and requires an explicit opt-in header. The company advises users to evaluate the feature for their own use cases.
What image files does Jev accept?
The documented limits cover PNG and JPEG still images, with each side between 256 and 1,568 pixels. Each image can be up to 16 MiB, and the total request limit is 32 MiB. WebP and HEIC files must be converted first.
Source: ThorstenMeyerAI.com
