🔍 Read the full analysis: I Pointed Jev At My Images To Explore Its AI Vision on ThorstenMeyerAI.com
Get the latest gadgets 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 answer type. In a test described by Thorsten Meyer, Jev matched product images to their headlines well, but results across news images were less straightforward; the source account is cut off before those findings are fully reported.
Jev is designed to return structured decisions rather than write prose: it can answer yes-or-no questions as probabilities, choose among options, or score a result. Before this release, those questions were limited to text. TypeSafe says image input also works with jev-1.13.0, which the company described as having been multimodal already, although that capability had not been exposed in the same way.
Image processing requires the request header TypeSafe-Danger-Zone-Images: true. In Jev 1.14, developers can also request one bounding box for a specified object, with a confidence score. That feature uses the box type and is not available in Jev 1.13. Images can be placed in the request state, question instructions, or criteria used to define answer options and score levels.
Meyer tested the system on 30 recent featured images from ThorstenMeyerAI.com: 10 product roundups, 15 news posts and five essays. He compared whether each image fit its own headline, a randomly chosen different headline, and an unrelated houseplant headline. He also asked the model to identify the product in each of the 10 product images, running tests on both model versions for about 60 requests in total.
For product images, Meyer reported that Jev 1.14 assigned the correct headline scores from 0.65 to 0.89, with a median of 0.76. Random mismatched headlines scored no higher than 0.18, and the unrelated plant headline no higher than 0.10. Jev 1.14 identified the product correctly in nine of 10 images. The source account says Jev 1.13 got all 10 product identifications right, but gave two answers confidence scores 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
Experimental Vision With Guardrails
The release extends Jev from text-based decisions to image-based inputs, potentially allowing developers to apply the same structured questions to visual material. Examples include checking whether an image suits a headline, identifying an object, or selecting among image-based options. The bounding-box output adds a more specific way to locate a target, though TypeSafe has not endorsed it for production use.
The explicit opt-in approach matters because it keeps existing requests on their prior behavior unless developers deliberately enable the experimental functions. TypeSafe says there is still too much “jaggedness” to recommend the new features for production. Meyer’s small, site-specific test offers an initial example, not an independent benchmark or broad evidence that the model will perform similarly on other image collections.
Cost and input handling may also affect adoption. TypeSafe says images are billed as ordinary input tokens at Jev’s stated rate of $0.042 per million input tokens, with no separate image surcharge. However, image placement can affect repeated processing, and the service applies file and request size limits.
As an affiliate, we earn on qualifying purchases.
How Jev’s Image Requests Work
TypeSafe’s image method uses an assets map that associates a reference string with image data. The reference can then appear in the state or in question instructions, where Jev substitutes the corresponding image. The source account recommends using a fresh UUID for each reference: matching is plain text, so a short string could unintentionally replace part of another word.
Placement has practical consequences. An image included in the shared state is used across the questions in a request; an image placed in an individual question’s instructions is sent again for each question that uses it. TypeSafe’s stated limits are PNG or JPEG still images, 256 to 1,568 pixels per side, up to 16 MiB per file and 32 MiB per request. WebP and HEIC files must be converted, and a single invalid image can cause the whole request to fail.
The same release includes a separate opt-in for longer context. Requests normally allow up to 64,000 tokens, with 32,000 per question; the header raises both limits to 262,000. TypeSafe retains the lower default and advises against longer inputs because reliability can decline. The company also says developers can opt in to sharing queries to help identify weaknesses, while stating that it will not use those queries to train the model.
TypeSafe released Jev 1.14 as jev-preview. It said jev-latest would move to the new version about a week after release, around October 21, 2026. The company characterized the model update as a set of broad, modest improvements and said multilingual support was expected to benefit most. Its comparisons used internal estimates against open-weight models, not public benchmark scores, and TypeSafe warned that those estimates might favor Jev.
““There is still way too much jaggedness for us to vouch for these for production use cases.””
— TypeSafe, describing the experimental features in its release post, as quoted by Thorsten Meyer
As an affiliate, we earn on qualifying purchases.
Results Beyond Product Photos
The source article supplied for this report ends partway through its account of the 15 news-image tests, at the sentence “News images: Jev was right, my images weren’t.” It does not provide the remaining results or explain that wording, so the news-image findings cannot be assessed here. Results for the five essays are also not included in the supplied material.
Meyer’s evaluation covers images from one website and uses a limited set of questions. It does not establish how Jev performs across other industries, image types, or more demanding visual tasks. TypeSafe itself has not vouched for the capabilities in production. Details such as performance on varied image sizes and the reliability of bounding-box results are not reported in the supplied account.
The release description also does not provide public benchmark results for Jev 1.14. TypeSafe’s stated internal comparisons are limited to open-weight models, and the company acknowledges possible bias in its estimates. No independent evaluation is included in the source material.
As an affiliate, we earn on qualifying purchases.
Jev Preview And Developer Testing
Developers can try Jev 1.14 through jev-preview, enabling image support or longer context through their respective headers. TypeSafe said jev-latest was expected to point to Jev 1.14 around October 21, one week after release. Those dates describe the company’s rollout plan, not confirmation in the source that the change has already taken place.
TypeSafe said Jev 1.13.0 would receive 60 days’ deprecation notice and Jev 1.14.0 30 days, with a formal policy still to come. For now, the clearest next step for teams considering the vision features is to test them against their own images and use cases, while treating the outputs as experimental rather than production-validated.
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 bounding-box response type to Jev 1.14. Image support also works with Jev 1.13.0, according to the company.
Are Jev’s image features ready for production?
TypeSafe says they are not ready for production use and requires an explicit danger-zone header to enable image support. The box feature is available in Jev 1.14.
How did Jev perform in Meyer’s image test?
In the reported product-photo tests, Jev 1.14 matched images to their own headlines with scores from 0.65 to 0.89 and identified nine of 10 products correctly. The supplied source does not include the complete news-image findings or results for the essays.
What image files can developers submit?
The stated limits cover PNG and JPEG still images, between 256 and 1,568 pixels per side, up to 16 MiB per image and 32 MiB per request. WebP and HEIC files need conversion first.
When does Jev 1.14 become the latest version?
TypeSafe said jev-latest would move to Jev 1.14 about a week after its October 14 release, around October 21, 2026. The supplied report does not confirm whether that rollout has happened.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
