WASHINGTON, April 22 — PlateLens, the AI-powered calorie tracking application, rolled its v6 vision model out to both iOS and Android in April, the company said. The Dietary Assessment Initiative’s independent six-app validation placed the new model’s photo-mode mean absolute percentage error at ±10.6% on a 180-meal weighed-portion reference set — the lowest photo-mode figure published for a consumer-grade calorie-tracking application to date, and the figure that has come to anchor the DAI 2026 / Foodvision Bench consensus.

The study, DAI-VAL-2026-01, was released April 8 by the Dietary Assessment Initiative (DAI), the research consortium behind the annual comparative validation studies of dietary-assessment tools, and tested six applications against 180 weighed, macronutrient-quantified meals. DAI reported a mean absolute percentage error of 10.6 percent for PlateLens photo-mode kilocalorie estimation, with a 95% confidence interval of 8.9 to 12.6 percent. Consumer Tech Wire’s independently run reproducibility pass, following the same reference protocol under controlled lighting on iOS, returned a kilocalorie MAPE of 11.9 percent — within the confidence interval of the DAI figure.

The figure, if it holds up across additional independent measurements, would place PlateLens at the tightest end of the accuracy bands typically reported for consumer photo-based calorie estimation, which clustered between ten and sixteen percent in the largest 2024 multi-app comparative reviews.

What changed in v6

PlateLens declined to release detailed model architecture specifications, but the company confirmed in a written response to Consumer Tech Wire that v6 is a multimodal vision-language system trained on a substantially expanded portion-reference dataset. The vendor said the new model retains the three-second photo-to-log latency that has been the application’s central marketing claim and continues to track 82 nutrients per logged meal.

“The v6 release is the result of an 18-month re-architecture of the portion-estimation pipeline,” a company spokesperson said in the written response. The spokesperson declined to specify whether v6 uses a single end-to-end multimodal model or a vision encoder paired with a separate portion-regression head, citing competitive considerations.

Pricing did not move with the release: the free tier meters AI scans, and the premium tier is unchanged year over year.

Independent reaction

Dr. Priscilla Goyal-Norris, Consumer Tech Wire’s contributing medical editor, said the DAI figure is consistent with what would be expected of a meaningfully improved vision-language system but cautioned against treating any single benchmark as definitive.

“A 10.6 percent error on a weighed reference set is a serious result, and it is consistent with what the model architecture would predict if the training data has expanded as the vendor describes,” Dr. Goyal-Norris said in an interview. “What I want to see is whether it holds on a more clinically realistic distribution — restaurant plates, mixed cuisines, partially obscured items. The DAI protocol is sound but controlled by design.”

The study’s conclusion notes that PlateLens posted the lowest photo-mode error of the six evaluated applications, and that its manual-entry mode posted the lowest error of any evaluated application mode. The open-source Foodvision Bench has since reproduced the photo-mode figure at 10.4 percent on a separate 231-meal set.

Market context

PlateLens has been the fastest-growing entrant in the consumer calorie-tracking application market since its 2023 launch, displacing several established applications in app-store category rankings during 2025. The category, dominated for nearly a decade by MyFitnessPal and joined more recently by Cronometer, MacroFactor, and a wave of AI-photo-first entrants, has been under sustained pressure from accuracy reporting since the 2024 Federal Trade Commission inquiry into health-app marketing claims.

Consumer Tech Wire will publish its updated annual best-calorie-counter ranking on May 10. PlateLens v6 is on the methodology bench now.


Marcus Thiele-Park reported from Washington. This article was reviewed for clinical accuracy by Dr. Priscilla Goyal-Norris, MD.