🔍 Read the full analysis: Jev Can See Now: I Tested It On 30 Of My Own Images on ThorstenMeyerAI.com
Get movie nights 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 inputs and a bounding-box answer type in Jev 1.14, released October 14, 2026. Thorsten Meyer tested the image features on 30 images from his site; product images matched their headlines well, while the supplied report cuts off before giving results for news images, essays or the full model comparison.
Thorsten Meyer tested TypeSafe’s new Jev image features on 30 featured images from ThorstenMeyerAI.com, reporting strong headline matching for product photos but not providing complete results for the other image groups in the available account. The test followed Jev 1.14’s October 14 release, which added opt-in image input and a bounding-box response type to a model previously framed around text-based decisions.
Meyer says he submitted 30 recent images from his site: 10 product roundups, 15 news posts and five essays. For each image, he tested whether it fit its own headline, a randomly selected different headline and an unrelated headline about repotting a houseplant. Product images also received a multiple-choice question asking which of 10 products the image showed. He reports running the tests on Jev 1.14 and 1.13, with about 60 requests overall.
For the 10 product images, Meyer reports that Jev 1.14 scored each image’s match with its own headline from 0.65 to 0.89, with a median of 0.76. A randomly paired headline scored no higher than 0.18, with a median of 0.06; the unrelated plant headline scored no higher than 0.10. Jev 1.14 named the right product in nine of the 10 images. The miss involved a mesh Wi-Fi photo showing both a router with antennas and a cylindrical mesh unit; the model selected “Wi-Fi 7 router” with 0.60 confidence.
Meyer says Jev 1.13 identified all 10 products correctly, though two answers had confidence below 0.5. He notes that all of Jev 1.14’s correct product identifications had confidence of at least 0.77, suggesting a threshold might have flagged the incorrect answer in this small test. The source excerpt then begins a section on news images but ends before reporting its findings, so it does not establish how the models performed across all 30 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
Where Image Decisions May Help
The release extends Jev from structured decisions about text to questions involving visual inputs. That could support tasks such as checking whether an image fits a headline or locating a specified object, while keeping the output in a format that software can evaluate: a probability, choice, score or bounding box.
Meyer’s product-image results offer an initial example, not evidence of broad reliability. The test used one publisher’s images and a particular set of prompts, and the excerpt reports detailed findings only for 10 product photos. For readers considering the model, the key practical point is TypeSafe’s own caution: the image features are experimental, and the company says they are not ready for production use without further validation.
As an affiliate, we earn on qualifying purchases.
Jev Adds Opt-In Image Inputs
Jev is described by Meyer as a decision model that does not write free-form sentences. Users supply a state and typed questions, and the model responds with structured answers, such as yes or no with a probability, a selection from a list, or a score with confidence. Before this release, the questions in that setup had to concern text, according to his account.
TypeSafe’s October 14 announcement added image support for jev-1.13.0 and jev-1.14.0, plus a new box type for Jev 1.14 that returns one bounding box and a confidence for a described target. The company also raised the maximum context length through a separate opt-in header, while retaining the previous default limits. Meyer says TypeSafe warned that reliability declines on long inputs and characterized the new capabilities as too jagged to endorse for production use.
Images are attached through an assets map and referenced in the request’s text. According to the source, supported still-image formats are PNG and JPEG, with limits of 256 to 1,568 pixels per side, 16 MB per image and 32 MB per request. The account says images are billed as normal input tokens at Jev’s stated rate, with no separate image surcharge.
““A confidence threshold would have caught the miss.””
— Thorsten Meyer
As an affiliate, we earn on qualifying purchases.
Results Beyond Product Photos
The supplied source text is incomplete: it stops as Meyer begins discussing news images. It gives no results for the 15 news images, the five essays, the randomly paired and off-topic headline tests across those groups, or the complete comparison between Jev 1.14 and 1.13. Those missing findings prevent a full assessment of the 30-image test.
The reported figures also come from a single site’s images and prompts. The source does not describe independent review, repeat runs, or a benchmark against human judgments. TypeSafe itself cautioned against relying on the experimental features in production. The results therefore should be treated as a limited test reported by the publisher, not as a general measure of Jev’s image accuracy.
As an affiliate, we earn on qualifying purchases.
Release Rollout and Further Testing
TypeSafe made Jev 1.14 available as jev-preview, with the jev-latest alias scheduled to move to the new version about a week after release, around October 21, according to Meyer’s account. The company said it would give 60 days’ notice before deprecating version 1.13 and 30 days for version 1.14, while a formal deprecation policy was still to come.
For users exploring image decisions, the immediate next step is local testing against their own images and prompts, while keeping the experimental features opt-in. The full findings from Meyer’s news-image and essay tests are not present in the supplied excerpt, and no broader production endorsement or independent evaluation is reported there.
image bounding box annotation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What changed in Jev 1.14?
TypeSafe added opt-in image inputs and a new bounding-box response type to Jev 1.14. Image support also works with Jev 1.13, according to the release details reported by Meyer.
What did the 30-image test find?
The available account reports detailed findings for 10 product images: Jev 1.14 matched images to their own headlines with a median score of 0.76 and identified nine of 10 products correctly. The excerpt does not include the results for the news images or essays.
Is TypeSafe recommending Jev’s image features for production?
No. Meyer quotes TypeSafe as saying the new capabilities have too much “jaggedness” for the company to vouch for production use. The features require an explicit opt-in header.
When does Jev 1.14 become the latest version?
According to Meyer’s report, Jev 1.14 was available as jev-preview after its October 14, 2026 release. TypeSafe planned to move jev-latest to it about one week later, around October 21.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
