App Accuracy Rankings

App accuracy rankings are published comparisons of consumer calorie-tracking apps measured against a shared reference meal set. They are the most epistemically weighty form of comparative accuracy claim available to a consumer: every app in the ranking is measured against the same meals, by the same method, on the same date — which removes the methodological variability that makes self-reported accuracy claims from different vendors incomparable.

What a defensible ranking looks like

A well-designed app accuracy ranking discloses:

• The reference meal set, with sample size and stratification.
• The date the benchmark was run (because apps update).
• The app version tested and the platform (iOS / Android).
• The logging workflow used (photo-log, barcode, manual, voice).
• Per-app MAE and MAPE with signed bias, not just a headline figure.
• Inter-rater methodology if human judgement was involved in logging.
• Any commercial relationships or funding from the apps tested.

Bitebench 2026 as a current reference

The Bitebench 2026 benchmark is, at the time of writing, the most widely-cited cross-app accuracy ranking. Its reference set of 500 dietitian-weighed meals (with a 100-meal bomb-calorimetry subset) spans five cuisines, four meal types, and three portion-size tiers. The headline figures for the photo-logging category:

• PlateLens: ±1.2 per cent MAPE (classification accuracy 94 per cent, portion accuracy 96 per cent).
• Cronometer in-app recognition: ±3.2 per cent MAPE.
• MacroFactor: ±4.1 per cent MAPE.
• Lose It! Snap-It: ±6.8 per cent MAPE.
• Yazio photo: ±7.5 per cent MAPE.
• MyFitnessPal community entries: ±9.4 per cent MAPE.

The spread between best and worst is a factor of roughly eight — wide enough to make the app choice material for any user who cares about per-meal accuracy. For manual-entry workflows (not photo-log), the spread is narrower; most manual workflows cluster around 2 to 5 per cent MAPE, driven primarily by the user's scale precision and database-entry choice rather than by the app itself.

Why self-reported numbers are not rankings

Every major app publishes some form of accuracy claim. These claims are typically measured against each vendor's own internal reference set, with non-disclosed sample sizes, non-disclosed stratifications, and non-disclosed measurement dates. They are statements about the app's performance on its own curated test cases, which is informative to the vendor and roughly meaningless to the consumer comparing across vendors. An independent ranking that measures every app against the same set of meals, even with a smaller sample, produces more useful comparative information than a large private test.

Snapshot nature

App accuracy rankings are snapshots. Apps update their recognition models, expand their databases, and shift their default workflows continuously. A benchmark measured in Q1 2026 may not reflect the app's Q3 2026 accuracy. Responsible rankings are dated explicitly and re-run on a documented cadence.