also searched as: is cal ai accurate · how accurate is cal ai · cal ai accuracy · cal ai accurate · how accurate is cal ai app · is cal ai app accurate · is cal ai reliable
Kinda: in the one independent test that includes it, Cal AI undercounted weighed meals by about 345 calories each on average. The test is preliminary, and most of the missing calories were fat a camera cannot see.
Trust it for a rough picture of your week, like which meals are heavy and how often you snack, where the pattern matters more than the count.
Double-check it when the meal was cooked in oil or butter, has a sauce, or is a mixed dish like a curry, because that is where photo estimates miss most.
Don't rely on it if your diet was set by a doctor, or calorie numbers have fed anxiety or disordered eating for you: a food scale and a dietitian are the tools, not a photo.
sources checked 2026-10-13 · by dan scutaru
| who tested it | what they found |
|---|---|
| NIH researchers (NIDDK), July 2026 | 345 calories low per meal Cal AI's average undercount (95% CI 296 to 392) on photos of 102 meals weighed to 0.1 g in a metabolic kitchen; a conference abstract, not yet peer reviewed. source |
| Same NIH test, all four apps | 250 to 345 calories low, fat 30 g low MyFitnessPal, Lose It!, Cal AI and Appediet all ran low on calories and fat, carbs held up best, and an early look at 200 more meals found ketogenic meals worse. source |
| University of Sydney, Nutrients, 2024 | mixed dishes off by up to 270% Seven photo apps, Cal AI not among them, on 22 food and drink photos: mixed-dish errors ran both ways, and butter and oil added in cooking went undetected. source |
| Cal AI's own claim | "about 80% accurate" Stated in Cal AI's FAQ (last updated July 2026), with no method or test set behind it. source |
About 345 calories low per meal, in the one controlled test we found. Researchers at NIDDK, part of the US National Institutes of Health, photographed 102 meals with every ingredient weighed and ran the photos through Cal AI, MyFitnessPal, Lose It! and Appediet. All four undercounted, by about a third on average. Cal AI's average gap was the largest, with MyFitnessPal (327) and Lose It! (333) close behind. The work was presented at the American Society for Nutrition's July meeting and is not yet peer reviewed.
From the surface. Cal AI's homepage says your phone's depth sensor calculates food volume, and the model names what it sees. Volume is not calories. A camera cannot see the oil vegetables were fried in, the butter in the rice, the cream in a sauce, or how deep the bowl is. Those hidden extras are mostly fat, which fits the NIH finding that fat was undercounted by about 30 g a meal.
Not accurate enough to trust one meal at a time, and Cal AI is not an outlier. In the 2024 Sydney study, only four of seven apps estimated calories from the photo at all, and on mixed dishes errors ran from 55% under on a pearl milk tea to 270% over on bibimbap. Single foods were not safe either: average errors ran from 47% under to 44% over, depending on the app.
Yes, as a fast way to log: take a photo, get calories and macros, once you subscribe. Whether the number is right is a separate question, and the one controlled test found it about 345 calories low per meal.
Only if a rough number that runs low still helps you, because you pay for speed, not precision. What it costs is on our Is Cal AI free? page.
MyFitnessPal, which announced the acquisition on March 2, 2026 and kept Cal AI as a standalone app. MyFitnessPal's own photo logging was in the same NIH test and ran about 327 calories low per meal.
Yes, it is a real app with a 4.8 rating from about 370,000 App Store ratings. Legit and accurate are different questions: the company says about 80%, and the only independent test found it reads low.
No. Its App Store listing says it does not offer medical advice, and its terms call estimates approximations and disclaim liability for wrong counts. For a diet a doctor set, weigh your food and work with a dietitian.