30 Of My Images Put Jev’s New Vision To The Test

🔍 Read the full analysis: 30 Of My Images Put Jev’s New Vision To The Test on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

Thorsten Meyer tested TypeSafe’s Jev 1.14 decision model on 30 featured images from his website after the company added image input and a bounding-box response type on October 14, 2026. In the results available from Meyer’s report, the model matched product photos to their own headlines and correctly identified 9 of 10 products, while TypeSafe cautioned that the new capabilities are not ready for production use.

Meyer’s test used 30 recent images from ThorstenMeyerAI.com: 10 product roundups, 15 news posts and five essays. He asked whether each image fit its own headline, a randomly selected other headline, and an unrelated headline about repotting a monstera plant. The 10 product images also received a multiple-choice question asking which product each showed. Meyer says he ran the test on both Jev 1.14.0 and 1.13.0, using about 60 requests in total.

For product roundups, Meyer reports that Jev 1.14 gave each image’s own headline a score from 0.65 to 0.89, with a median of 0.76. Random alternative headlines reached no higher than 0.18, with a median of 0.06; the unrelated plant headline scored no higher than 0.10. Jev 1.14 selected the pictured product correctly in nine of 10 cases. Its miss was a mesh Wi-Fi image: the model called it a “Wi-Fi 7 router,” with confidence of 0.60.

Meyer says every correct product identification from Jev 1.14 had confidence of at least 0.77, while Jev 1.13 identified all 10 products but assigned confidence below 0.5 to two answers. The supplied source ends as Meyer begins describing the news-image results. It therefore does not provide outcomes for those 15 images or the five essays, and no conclusions about those parts of the test can be drawn from the available material.

What Image Inputs Add to Jev

The test offers an early, small-scale example of how a structured decision model might assess visual material. Rather than generating a caption or a paragraph, Jev answers predefined questions with a probability, choice or score. For publishers, a headline-fit check could help flag images that appear unrelated to the story, though Meyer’s results do not establish how reliably that would work beyond his sample.

The product results also illustrate why confidence scores need interpretation. Jev 1.14’s one incorrect product answer had lower confidence than its correct answers in this test, which suggests a threshold might have flagged it. But 10 product images are not enough to validate a dependable threshold, and Meyer reports no broader accuracy study. TypeSafe itself says the capabilities remain too unreliable for production use without further evaluation.

Jev’s Opt-In Image Release

Before this release, Jev’s questions had to concern text, according to Meyer’s account of the model. TypeSafe’s October 14 update added image support to Jev 1.13.0 and 1.14.0, activated only when a request includes an explicit danger-zone header. The release also introduced a box response type in 1.14.0, which returns one bounding box and a confidence for a described target.

TypeSafe says images can be placed in a request’s state, question instructions or criteria. The company also added an opt-in for longer context: the standard limits remain 64,000 tokens per request and 32,000 per question, while the header raises each limit to 262,000. TypeSafe advises against the larger context because reliability falls on long inputs. Jev 1.14.0 is available as jev-preview; Meyer’s source says jev-latest was scheduled to move to it about a week after release.

“The results were more honest about my images than I expected.”

— Thorsten Meyer

Results Missing for 20 Images

The source text provided for this report stops midway through the section on news-image results. It does not state how Jev performed on the 15 news images or five essays, nor does it give an aggregate score across all 30 images. The reported product outcomes are Meyer’s test results, not an independent evaluation or a large-scale benchmark.

Other details also remain unsettled. The report says TypeSafe does not recommend production use of the new features, and it does not establish how performance changes across different image types, headline styles or datasets. TypeSafe’s internal estimates are described by the company as potentially biased toward Jev; the source does not provide public benchmark scores to resolve that concern.

Further Testing and Model Rollout

TypeSafe’s release plan, as described by Meyer, was to move jev-latest to Jev 1.14.0 about one week after the October 14 release, around October 21. The company also said it would give 60 days’ deprecation notice for version 1.13.0 and 30 days for 1.14.0, with a formal policy still to come. Those dates and policy details are based on the October 15 report and may require confirmation as the rollout proceeds.

For users considering image-based workflows, TypeSafe’s stated position is to opt in cautiously and test their own use cases. Meyer’s partial results show what the new model did on a small group of product images, but the missing results and TypeSafe’s reliability warning leave open whether it will perform consistently across broader production material.

Key Questions

What did Jev add in version 1.14?

TypeSafe added image input and a new bounding-box response type. Image support also works with Jev 1.13.0, according to TypeSafe as quoted in Meyer’s report.

How did Jev 1.14 perform on the product images?

Meyer reports that it matched each product image to its own headline with scores from 0.65 to 0.89 and identified the pictured product correctly in nine of 10 cases. These are results from his limited test, not a general benchmark.

Did the test cover all 30 images?

Meyer says the test included 10 product images, 15 news images and five essays. But the source text supplied here cuts off during the news-image section, so the results for the news images and essays are not available.

Is TypeSafe recommending Jev’s image features for production?

No. TypeSafe says the features remain too unreliable for it to vouch for production use and places them behind an explicit opt-in header.

Source: ThorstenMeyerAI.com