🔍 Read the full analysis: Jev Can See Now: What 30 Images Revealed In My Test on ThorstenMeyerAI.com
Get audio and creator gear 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 model on 30 images from his site; the available results show strong product-image matching, but the source ends before it reports the news and essay findings.
TypeSafe’s Jev decision model gained opt-in image input and a new bounding-box output with the October 14 release of Jev 1.14. In a test published the next day, Thorsten Meyer tried the image feature on 30 images from his website; the reported product-image results were strong, though the supplied account ends before it gives results for news and essay images.
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. TypeSafe’s release makes images available in the state, question instructions or criteria used to define options and scoring levels. Image input also works with Jev 1.13, TypeSafe said. The new box question type, which returns one bounding box and a confidence score for a described target, is limited to Jev 1.14.
Both image input and longer-context requests are guarded by explicit headers. TypeSafe said it is not ready to recommend the new capabilities for production, describing them as too inconsistent for that use. Image requests accept PNG and JPEG stills within set size limits; the source says a single invalid image can cause the entire request to fail. TypeSafe also introduced an opt-in setting allowing it to use submitted queries to identify model weaknesses, while saying it will not train on those queries.
Meyer tested the 30 most recent featured images on ThorstenMeyerAI.com: 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 houseplant headline. The 10 product images also received a question asking Jev to identify the product shown. The test used both Jev 1.14 and 1.13, with about 60 requests in total, according to Meyer.
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
What the Product Test Shows
The reported results suggest Jev can perform a narrow visual-matching task, but they do not establish how well it works across image types or in production. For the 10 product images, Jev 1.14 gave each image a higher score for its own headline than for a random other headline: own-headline scores ranged from 0.65 to 0.89, with a median of 0.76. Random-headline scores reached no higher than 0.18, with a median of 0.06. The unrelated plant headline scored at most 0.10.
On product identification, Jev 1.14 named the correct item in nine of 10 cases. Meyer described the one miss as understandable: a photo showed a router beside a cylindrical mesh unit, and the model chose “Wi-Fi 7 router” with 0.60 confidence. He reported that the correct answers had confidence scores of at least 0.77, which suggests a threshold could have flagged that error in this small test. Jev 1.13 identified all 10 products correctly, but gave two answers confidence below 0.5. These observations are useful as a test report, not a general performance guarantee.
The practical question for developers is whether image-based decisions can be made reliably enough for their particular workflow. TypeSafe’s opt-in requirement and production warning make clear that users should test against their own data and account for mistakes, rather than treating the feature as ready for unattended decisions.
As an affiliate, we earn on qualifying purchases.
How Jev Handles Images
Before this release, Jev’s typed questions operated on text. TypeSafe says image support was also present in Jev 1.13, though it had not been exposed for this use. The October 14 announcement made the capability available behind the TypeSafe-Danger-Zone-Images header, reflecting the company’s stated concern about reliability.
Requests refer to image files through an assets map. A reference string placed in the state or question instructions is replaced with the corresponding image. Meyer notes that references should be unique because matching is plain text; a short string could unintentionally match part of another word. Images placed in the shared state are used across questions, while an image inserted into individual instructions may be sent again for each question that uses it.
TypeSafe set a limit of 256 to 1,568 pixels per side, 16 MiB per file and 32 MiB per request, with PNG and JPEG supported. It says images are billed as ordinary input tokens at Jev’s stated rate of $0.042 per million input tokens, with output tokens free. The release also raises the optional context limit from 64,000 tokens per request and 32,000 per question to 262,000 for each, while TypeSafe advises against the longer setting because reliability can decline on long inputs.
“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
Results Beyond Product Images
The supplied source account cuts off as Meyer begins discussing the 15 news images, after the line “The news posts were a dif”. It does not provide findings for the news images or the five essays, so those results cannot be reported here. The account also does not give a complete breakdown of scores across both model versions or describe a controlled comparison against human judgments.
The reported test covers one website’s images and a limited set of questions. It does not establish performance on other image collections, image formats beyond the supported stills, or production workloads. TypeSafe itself warned that reliability remains uneven. The source also says the company shared internal intelligence estimates rather than public benchmark scores and cautioned that those estimates may favor Jev.
Jev 1.14 Rollout and Testing
TypeSafe made Jev 1.14 available as jev-preview on October 14 and said jev-latest would move to the new version one week after release, around October 21. The company stated that Jev 1.13 would receive 60 days’ deprecation notice and Jev 1.14 would receive 30 days, with a formal policy still to come.
For now, the main next step for teams interested in image questions is to run evaluations on their own images and tasks, as Meyer recommended. Users will need to opt in, observe the image and request limits, and decide how to handle low-confidence or incorrect answers. The source provides no further results from Meyer’s news and essay tests, and no later TypeSafe reliability update.
Key Questions
What changed in Jev 1.14?
TypeSafe added opt-in image input, a new box type that returns one bounding box and a confidence score, and broader model improvements. The box type is specific to Jev 1.14; image input also works with Jev 1.13, according to TypeSafe.
How did Jev perform on the 10 product images?
In Meyer’s test, Jev 1.14 scored each product image higher against its own headline than against a randomly selected other headline. It identified the product correctly in nine of 10 images. These are results from a small test on one website, not a broad benchmark.
Did the test cover news images and essays?
Meyer said the test included 15 news posts and five essays, but the supplied account cuts off before reporting those results. Their performance cannot be confirmed from the available source material.
Is TypeSafe recommending image features for production?
No. TypeSafe said the capabilities remain too inconsistent for it to vouch for production use and placed them behind an explicit opt-in header. The company advises users to evaluate the model for their own use cases.
When does Jev 1.14 become the latest version?
TypeSafe said Jev 1.14 was available as jev-preview from October 14, 2026, and that jev-latest would move to it about a week later, around October 21.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
