🔍 Read the full analysis: What 30 Images Taught Me About Jev’s AI Vision on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
TypeSafe’s October 14 release added opt-in image input and bounding-box output to Jev, a model designed to answer structured questions. Thorsten Meyer tested the new and previous versions on 30 images from his site; product-image results were strong, while the supplied report cuts off before giving the news-image findings or complete conclusions.
Jev is designed to respond to typed questions with a probability, a choice from a list or a score, rather than to compose prose. Before this release, its questions were text-based. TypeSafe says image support works with jev-1.13.0 and jev-1.14.0; users must activate it with the request header TypeSafe-Danger-Zone-Images: true. The release also introduces a box type, available only in Jev 1.14, which returns one bounding box for a specified target along with a confidence score.
Meyer tested whether images matched their own headlines, randomly selected other headlines and an unrelated plant-care headline. He used 30 featured images: 10 product roundups, 15 news posts and five essays. The 10 product images also went through a question asking Jev to identify which product each showed. Meyer says he ran the tests on both model versions, making about 60 requests in total.
For product images, Meyer reported that Jev 1.14 scored each image’s own headline between 0.65 and 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-care headline scored no higher than 0.10. Jev 1.14 identified nine of 10 products correctly. Jev 1.13 identified all 10, though two answers had confidence below 0.5. These are results from Meyer’s small, site-specific test, not an independent 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
A Cautious Route to Image Checks
The release gives developers a way to apply Jev’s structured decisions to visual inputs, including checks such as whether a featured image fits a headline or whether a target object appears in a picture. That could support repeatable review tasks where a probability or classification is more useful than a free-form description. Meyer’s product-photo results suggest this approach may be worth testing, but his sample is too small and narrow to establish general accuracy.
TypeSafe’s opt-in approach also sets a clear boundary around the feature. The company says the image and other new capabilities remain too inconsistent for it to endorse them for production use. Keeping them behind explicit headers means existing requests do not automatically change behavior, while developers can experiment and evaluate results against their own data.
As an affiliate, we earn on qualifying purchases.
How Jev’s Image Inputs Work
TypeSafe’s request format places image data in an assets map and inserts a reference wherever the image should be used—in the state, question instructions or criteria for options and scores. The company says a reference should be a fresh UUID: matching is plain text, so short references could accidentally replace part of another word. An image in the shared state can be used across questions; putting it in separate instructions may send it again for each question and affect cost.
The documented limits are PNG and JPEG still images, 256 to 1,568 pixels per side, with a 16 MiB limit per image and 32 MiB per request. WebP and HEIC files need conversion, and one invalid image can fail the request. TypeSafe says there is no image surcharge: input is billed at Jev’s standard rate of $0.042 per million tokens. It recommends multipart uploads because base64 encoding adds roughly one-third to the file size.
Jev 1.14 also raises the optional context limit from 64,000 tokens per request and 32,000 per question to 262,000 for each, if developers send a separate long-context header. TypeSafe advises against the larger limit because reliability can decline on long inputs. The company described its internal intelligence estimates as potentially biased toward Jev and recommended that users run their own evaluations.
“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 an affiliate, we earn on qualifying purchases.
The Missing News-Image Findings
The available source text ends as Meyer begins describing results for the 15 news images: it says, “The news posts were a dif” and provides no completed finding. It is therefore not possible to report how Jev performed on news images or on the five essays, or to describe Meyer’s full conclusions. The supplied material also does not include the images, request data or an independent replication.
The product results measure performance on one publisher’s selected images and headlines. They do not establish how the model would perform on other topics, image styles or production workloads. TypeSafe itself cautions that the new capabilities are not ready for production use, and the source gives no independent evaluation of the bounding-box feature.
As an affiliate, we earn on qualifying purchases.
Model Availability and Evaluation
TypeSafe said Jev 1.14 was available as jev-preview after the October 14 release. The company planned to move jev-latest to the new version about a week later, around October 21. It also outlined deprecation notice periods of 60 days for Jev 1.13.0 and 30 days for Jev 1.14.0, while saying a formal policy was still to come.
For developers, the immediate next step is to test the opt-in image and bounding-box features on relevant examples and check confidence scores against human-reviewed outcomes. The complete news-image results, which are absent from the supplied account, would be needed to understand how Meyer’s test performed beyond product roundups.
As an affiliate, we earn on qualifying purchases.
Key Questions
What changed in Jev 1.14?
TypeSafe added opt-in image inputs, a bounding-box response type for Jev 1.14, and an optional higher context limit. The company says these capabilities remain experimental and are not endorsed for production use.
What did Meyer’s 30-image test find?
In the product-photo portion, Meyer reported that 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 source provided here cuts off before the news-image findings.
Can Jev 1.13 analyze images?
According to TypeSafe’s release post, image support also works on jev-1.13.0, provided the image feature is enabled with the required opt-in header.
Is Jev’s image analysis ready for production?
TypeSafe says it is not yet ready to vouch for the new capabilities in production use cases. It requires explicit opt-in, and the company advises developers to evaluate performance themselves.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
