Disclosure: I develop Kaloro, which is number 10 on this list. It is placed last because it has the fewest ratings. The list is ordered by rating count, not by my opinion.
Packaged food and barcodes: MyFitnessPal or Lose It. Photo-first and most popular: Cal AI. Volume estimation done seriously: SnapCalorie. Home cooking, where every one of these apps comes in low: pick whichever app lets you correct the portion and the fat — which, as the scan below shows, is a very short list.
The model is not the problem. Every app here uses a capable vision model, and on a restaurant plate or a packaged snack they all land close enough. Where they all fail is home cooking, and they fail the same way and in the same direction: low.
That is not an opinion, it is a measured result. In a 2025 evaluation of three vision models against food weighed on a scale, ChatGPT and Claude landed at a mean absolute percentage error of 35.8% for energy, and — the important part — all models exhibited systematic underestimation that increased with portion size, with bias slopes ranging from −0.23 to −0.50
(Fridolfsson et al., Current Developments in Nutrition, 9 September 2025). A separate study of 114 meal photographs from national dietary survey data found ChatGPT-4 was good at naming the food — good precision (93.0%) for correctly identifying the foods in the photographs
— but poor agreement
for 10 of 16 nutrients, and it underestimated 11 of them, with agreement on weight fine for small meals and poor for medium and large ones (O'Hara et al., Nutrients, 7 February 2025).
So: these apps can tell chicken from tofu. What they cannot do is see how much of it there is, or what it was cooked in. Cooking fat is the most expensive invisible thing on the plate — fat carries about 9 kcal per gram (the standard factor in FAO's food-energy guidance), which makes a single 14 g tablespoon of olive oil worth about 126 kcal that no camera will ever record (FAO, Food energy: methods of analysis and conversion factors, 2003). Two tablespoons in a pan of menemen is a quarter of a kilogram of body weight a month if you never count it.
The fair comparison, by the way, is not perfection. Human self-reporting is also wrong, and also low: in a doubly-labelled-water study of 45 adults using an image-based food record, underreporting ran at 12% for men
and 10% for women
against measured energy expenditure (Boushey et al., Nutrients, 22 March 2017). Nobody in this category is offering you the truth. They are offering you an error bar, and the useful question is whether the app admits where its error comes from.
Three answers exist:
Nobody publishes this, so I measured it. On 24 September 2026 I pulled every app Apple returns for five calorie-scanner searches on the US App Store, kept the ones in a health or food category with at least 50 US ratings whose own description mentions both calories and photographing food — 79 apps — and read what each one claims.
| Claim in the app's own store description | Apps | What it tells you |
|---|---|---|
| Barcode scanner | 48 of 79 | The packaged-food problem is considered solved |
| You can edit, adjust or correct the result | 22 of 79 | Editing is common; editing what is the question |
| Mentions portion or serving size | 14 of 79 | Half as many as sell you a barcode scanner |
| Mentions home cooking or homemade food | 10 of 79 | The hardest case is the least advertised |
| Mentions cooking oil, butter or added fat at all | 2 of 79 | The single largest invisible error, named by almost nobody |
| Quotes an accuracy figure for its own scanner | 0 of 79 | Everyone is confident; nobody commits to a number |
The two apps that mention fat are worth separating. SnapCalorie is the only one that ties it to something you can do about it: its listing has a section headed COOKING OILS & ADJUSTMENTS
and tells you to add a quick voice note ("pan-fried in olive oil") or tap to adjust afterward
. Lose It! mentions butter only as an example of what you can say to its voice logger. Everyone else — including Cal AI, whose US store description does not contain the word "oil" — leaves the largest error on the plate unmentioned.
On the accuracy row: exactly one listing contained a percentage near the word accurate, and reading it showed a disclaimer rather than a claim — FitHub's listing says its AI suggestions aren't 100% accurate
. That is honest. It is also the only place in 79 store descriptions where a number appears at all. Cal AI does put one on its own website: CalAI is about 80% accurate. No food tracking app is perfect, and calories and nutrition values can vary.
(calai.app, read 24 September 2026.) That is a more candid sentence than most of this industry manages, and it is worth crediting.
| App | US ratings | What it does with the photo | Pick it if |
|---|---|---|---|
| MyFitnessPal | 2,363,049 · 4.7★ | Largest food database and barcode scanner; photo scan is a Premium add-on | You eat a lot of packaged food and want the biggest database |
| Lose It! | 778,192 · 4.8★ | Snap It photo logging on top of a big database | You want a classic tracker with photo as a bonus |
| Cal AI | 366,070 · 4.8★ | Photo-first, plus barcode and text description; long onboarding quiz before the paywall | You want the most popular photo-first app and do not mind a hard paywall |
| Lifesum | 150,729 · 4.6★ | Meal plans and diets with an AI photo option | You want diet plans more than a scanner |
| Yazio | 50,761 · 4.7★ | European database, fasting timer, photo scan | You are in Europe or Turkey and eat packaged food |
| Appediet | 47,199 · 4.8★ | AI photo counter with a meal-plan angle | You want a cheaper Cal AI-style app |
| Foodvisor | 17,554 · 4.6★ | One of the first photo counters; dietitian chat | You want human coaching alongside the scan |
| SnapCal | 12,577 · 4.8★ | Lightweight photo counter | You want something simple and fast |
| SnapCalorie | 6,647 · 4.7★ | Volume estimation plus the only cooking-oil adjustment I found in 79 listings | You care about portion volume and want to log the fat you cooked with |
| Kaloro (author's own app) | new, few ratings · —★ | Photo counter that shows its assumptions and lets you correct oil and portion; Turkish home cooking | You cook at home and the number keeps coming in low |
I read the 500 most recent public US reviews of Cal AI — the category leader, and therefore the best available sample of what goes wrong with a photo calorie counter — covering 30 May to 22 September 2026, on 24 September 2026. Of those, 253 were one or two stars. Counting by theme: 66 mentioned portion size, 49 used the word inaccurate or an equivalent, 41 described the onboarding questionnaire or paywall, 35 mentioned a refund or a cancellation problem — and only 2 mentioned oil, butter or what the food was cooked in.
That last number is the whole argument of this page. People feel the error constantly. Almost nobody can name its cause. The two who did, named it exactly:
"It's WAY low on calories with every meal. Remember it doesn't know what you cook with so all butter and oil is not accounted for. It's not just these guys, but conceptually…"1★, Cal AI, US App Store, 27 August 2026
"This app grossly underestimates the calories for meat unless you take a picture, then tell the app the weight of the meat and what you…"4★, Cal AI, US App Store, 12 August 2026
"This app is genuinely terrible at guessing portion sizes. What should be an 800-900 calorie meal says it's 400 every single time."1★, Cal AI, US App Store, 28 August 2026
And for balance, because a one-star sample is not the population — 161 of those 500 reviews were five stars, and they describe a product doing its job:
"This app has made it extremely easy for me a veteran to improve my health at 72. I need all the help I can get and Cal AI is excellent."5★, Cal AI, US App Store, 12 September 2026
Packaged food, barcodes, restaurant chains: MyFitnessPal or Lose It. The database matters more than the camera, and 48 of the 79 apps I scanned have a barcode scanner precisely because that part is easy.
Photo-first and you want the most popular option: Cal AI. Be aware that the onboarding is a long questionnaire that ends at a paywall; that theme appeared in 41 of the 500 recent US reviews I read.
You want portion volume taken seriously: SnapCalorie, which is also the only app in the scan that gives you a way to log the oil.
Europe or Turkey, mixed packaged and fresh food: Yazio, which has the best regional database. Its Turkish-language support is thinner than its database.
You cook at home and the numbers keep coming in low: Kaloro, or any app that lets you correct oil and portion rather than just the ingredient list. This is the case I built Kaloro for, so weigh that accordingly — and note that Kaloro has no barcode scanner and no food database, which the nine apps above all beat it on.
No photo can measure the oil you cooked with. Any app claiming a single accuracy percentage for "food photos" is quoting a benchmark built on plated, photographed, restaurant-style meals — and the research above shows the error grows precisely where that benchmark is weakest, on larger and home-cooked portions. On home cooking the honest answer is a range, and the app should show you which end of the range it picked.
Rating counts and averages are from the US App Store via Apple's public iTunes lookup API, read on 24 September 2026, rounded to one decimal. The 79-listing scan is my own: five search terms, filtered to health or food categories with at least 50 US ratings and a description mentioning both calories and photographing food, then matched against a fixed set of phrase patterns with every hit read by hand before counting — a first pass over-counted "assumption disclosure" by matching the phrase "estimates portions", and those rows were dropped rather than published. Review themes come from reading the 500 most recent public US reviews of Cal AI via Apple's customer-review feed; the counts are mine, not Apple's. Feature descriptions are from each app's own store listing. Nothing here was sponsored and there are no affiliate links. Corrections: necatidogrul7@gmail.com.