Can Jev Read The Details In My Pictures? I Tried 30

🔍 Read the full analysis: Can Jev Read The Details In My Pictures? I Tried 30 on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

TypeSafe added opt-in image input and a new bounding-box response type to Jev on October 14, and an initial test of 30 website images found that Jev 1.14 matched product photos to their headlines and identified 9 of 10 products correctly. The company cautioned that the new capabilities remain too unreliable for production use without further evaluation, making the results an early test rather than evidence of general performance.

TypeSafe’s Jev 1.14 release lets users provide images in the request’s state, a question’s instructions, or the criteria used to define possible answers. Image support also works with jev-1.13.0, according to TypeSafe. The new box response type, which returns one bounding box and a confidence score for a specified target, is limited to Jev 1.14.

The features are disabled by default and require explicit request headers. TypeSafe said it had not validated them for production use, citing “way too much jaggedness” in their behavior. Jev 1.14 also supports a longer context window through a separate opt-in header, though TypeSafe advises against the expanded limit because reliability can fall on long inputs.

Thorsten Meyer tested the two model versions on the site’s 30 most recent featured images: 10 product roundups, 15 news posts and five essays. Each image received questions about whether it fit its own headline, a randomly selected other headline and an unrelated plant-care headline. The product images also received a question asking which of 10 products each showed. The reported results available here cover the product images; the source text cuts off as it begins discussing the news-image findings.

What the Image Test Can Show

The test offers an early example of how an image-capable decision model might help with a narrow review task: checking whether a featured image fits a headline or identifying a product in a picture. In Meyer’s product-image sample, Jev 1.14 assigned each image’s own headline a higher fit score than the mismatched alternatives. That could make such a tool useful for screening or flagging items for human review, if broader testing supports the result.

But the test is small and site-specific. It does not establish how Jev performs across different image types, publishers, lighting conditions or more ambiguous editorial choices. TypeSafe’s warning that the feature is not production-ready also means the results should not be treated as a service guarantee. For readers and developers, the central development is the availability of image input—not proof that the model can reliably make editorial decisions at scale.

How Jev Added Image Input

Before this release, Jev was presented as a decision model that answers typed questions about supplied text. Its response formats include yes-or-no probabilities, selections from a list and scores on a scale, each with confidence information. TypeSafe described Jev 1.14 as a broad update, with expected gains in multilingual support, and made it available under the jev-preview model name.

Images are provided through an assets map: a reference string in the request is replaced with the image wherever that reference appears. TypeSafe’s documentation, as summarized by Meyer, limits uploads to PNG and JPEG still images from 256 to 1,568 pixels per side, with a 16 MB per-file limit and 32 MB per-request limit. WebP and HEIC files must be converted. TypeSafe says image inputs are billed as ordinary input tokens, with no separate image surcharge.

The release also introduced a longer-context option that raises the stated limits from 64,000 tokens per request and 32,000 per question to 262,000 for each, when enabled. TypeSafe said it retained the lower defaults because longer inputs can reduce reliability. The company also offers users an opt-in way to let it use queries to identify model weaknesses, while saying it will not train on those queries.

“jev-1.13.0 was multimodal all along.”

— TypeSafe, describing image support in its release post

Limits of the 30-Image Results

The available source material does not include the findings for the 15 news images or the five essays; it ends as Meyer begins that section. It is therefore not possible to report how Jev performed across the full 30-image set or to characterize the test’s overall accuracy.

The product results also come from one publisher’s images and headlines, with no independent replication or stated human-review benchmark. Meyer reported that Jev 1.14’s own-headline fit scores ranged from 0.65 to 0.89, with a median of 0.76; randomly paired headlines had a median of 0.06 and a maximum of 0.18, while the off-topic plant headline scored no higher than 0.10. These are observations from this sample, not general performance rates.

Jev 1.14 identified 9 of 10 products. Its one error was a mesh Wi-Fi image that showed both a router with antennas and a cylindrical mesh unit; the model chose “Wi-Fi 7 router” with 0.60 confidence. Meyer said all correct Jev 1.14 product identifications had confidence scores of at least 0.77, but a threshold’s performance would need testing on more examples. Jev 1.13 identified all 10 products, though two answers had confidence below 0.5.

Jev 1.14 Rollout and Testing

TypeSafe made Jev 1.14 available as jev-preview on October 14. The company said the jev-latest alias would move to the new version one week after release, around October 21, 2026. It also described deprecation notice periods of 60 days for Jev 1.13.0 and 30 days for Jev 1.14.0, with a formal policy still to come.

For now, TypeSafe’s stated position is that users should opt in deliberately and evaluate the model for their own tasks rather than assume production reliability. Further details about Meyer’s news-image and essay results, and independent tests of Jev’s image and bounding-box features, would help show how well the capabilities work beyond this initial product-photo sample.

Key Questions

What changed in Jev 1.14?

TypeSafe added opt-in image input, a bounding-box response type called box, and an option for longer context. The box type is Jev 1.14-only; TypeSafe says image support also works with Jev 1.13.0.

How did Jev perform on the product images?

In Meyer’s test, Jev 1.14 gave each of the 10 product images a higher fit score for its own headline than for the mismatched headlines. It identified 9 of 10 products. Jev 1.13 identified all 10, though two answers had confidence below 0.5.

Is Jev’s image support ready for production?

TypeSafe said the image and bounding-box features were not yet reliable enough for it to vouch for production use. They require an explicit opt-in header, and users should evaluate them against their own requirements.

What image formats and limits does Jev support?

The release details cited by Meyer specify PNG and JPEG still images, 256 to 1,568 pixels per side, up to 16 MB per image and 32 MB per request. WebP and HEIC files need conversion before use.

Are results available for all 30 images?

No. The source material reports the product-image findings but cuts off as the news-image results begin. It does not provide findings for the 15 news posts or five essays, so the full test outcome is unknown.

Source: ThorstenMeyerAI.com