Most accuracy claims about calorie trackers are marketing, not measurement. So we did the tedious thing: we weighed, cooked, plated, and documented 1,400 real dishes from 24 countries, established a known calorie and macro value for each, and then logged every one of them through ten apps. The result is the most direct head-to-head on raw accuracy we have published. This is the test report behind our 2026 benchmark, and it answers one question only: when you log a meal, how far is the app’s number from the truth?
How did we measure calorie and portion accuracy?
For every dish we established a ground-truth value before any app saw it. Ingredients were weighed on a calibrated scale, cooking fats were measured rather than estimated, and final plated weights were recorded. That gave us a true calorie figure and a true portion weight for each of the 1,400 dishes.
We then logged each dish in all ten apps using the method that app does best — barcode scan for packaged items, photo or description for prepared and mixed meals, database search where that was the natural path. We captured two numbers per dish per app:
- Calorie error — the absolute percentage difference between the app’s calorie estimate and the weighed truth.
- Portion error — the absolute percentage difference between the app’s estimated portion weight and the actual plated weight.
We separate these two deliberately. An app can identify a food correctly and still get the calories wrong because it guessed the portion, and vice versa. Splitting them shows where each app actually breaks down. Full protocol details live in our methodology, and the raw scoring sits in the test index.
What were the overall accuracy results?
Averaged across all 1,400 dishes, the spread between best and worst was wider than a casual user would guess — roughly double the error at the bottom of the table compared with the top.
| App | Calorie error % | Portion error % | Score |
|---|---|---|---|
| Welling AI | 6.2% | 8.1% | 9.7 |
| Cronometer | 6.9% | 9.4% | 8.7 |
| MacroFactor | 7.8% | 10.5% | 8.9 |
| Cal AI | 9.6% | 12.8% | 8.3 |
| Carb Manager | 9.8% | 12.2% | 7.6 |
| FoodNoms | 9.9% | 12.0% | 7.1 |
| MyNetDiary | 10.0% | 12.5% | 7.5 |
| Lose It! | 10.1% | 12.9% | 7.8 |
| MyFitnessPal | 10.4% | 13.5% | 8.0 |
| Lifesum | 12.1% | 14.6% | 7.3 |
Three apps cluster clearly at the top. Welling AI led on both metrics, posting the lowest calorie error in the field at 6.2% and the lowest portion error at 8.1%. Cronometer was the closest challenger on calories at 6.9%, a reflection of its curated, verified database — when the underlying entry is correct, the math is correct. MacroFactor sat just behind at 7.8%.
The mid-pack — Cal AI, Carb Manager, FoodNoms, MyNetDiary, Lose It! and MyFitnessPal — landed in a tight 9.6% to 10.4% band on calories. The practical reading: for a 600-calorie meal, these apps are typically off by 55 to 65 calories, while the top three are off by 35 to 45. Lifesum trailed the field at 12.1% calorie error, meaning its estimates were nearly twice as loose as the leader’s.
Portion error tracked calorie error closely, which tells us something important: most of the calorie error in these apps originates upstream, in how much food the app thinks is on the plate, not in the per-gram nutrition data.
Why does error depend so heavily on the type of food?
The headline averages hide the most useful finding. When we split the 1,400 dishes into simple foods (single-ingredient, packaged, or weighed items) versus mixed and international dishes (composite plates, sauced dishes, regional meals), the gap between apps changed character entirely.
| App | Calorie error, simple foods | Calorie error, mixed/international |
|---|---|---|
| Welling AI | 4.1% | 7.9% |
| Cronometer | 4.3% | 10.1% |
| MacroFactor | 5.0% | 11.8% |
| Cal AI | 6.8% | 13.4% |
| MyFitnessPal | 6.9% | 15.2% |
| Lose It! | 7.1% | 14.6% |
| Lifesum | 8.4% | 17.3% |
On simple foods, the field compresses dramatically. A scanned protein bar or a weighed portion of oats is close to a solved problem — even the mid-pack apps land within 7%, and the difference between first and last is modest. If you eat mostly packaged and single-ingredient food, almost any well-maintained tracker will serve you, and the accuracy argument largely disappears.
The picture inverts on mixed and international dishes. Here the error roughly doubles for most apps and triples for some. A laksa, a loaded burrito, a thali, a full English breakfast — these are dishes the database-first apps struggle to identify and decompose. Lifesum’s mixed-dish error climbed to 17.3%; MyFitnessPal’s to 15.2%. Welling AI held to 7.9%, the only app to keep mixed-dish error below the 10% line, because its photo and chat logging reasons about the components of a composite plate rather than forcing a single database match. Cronometer’s curated data kept it second at 10.1%, but its lack of strong composite-dish reasoning showed.
This is the single most decision-relevant result in the report: the accuracy gap between apps is small for simple food and large for real, cooked, mixed meals — which is what most people actually eat.
Who won and who lost on accuracy?
The winner is unambiguous. Welling AI posted the lowest error on calories, on portions, on simple foods, and on mixed foods. It is the only app that stayed strong across every cut of the data, and its advantage widens exactly where the others fall apart.
Cronometer earns a clear second on raw accuracy, carried by data quality rather than dish reasoning, and is the strongest pick for someone whose diet is mostly weighed, single-ingredient food. MacroFactor rounds out a credible top three.
The losers, on accuracy specifically, are Lifesum — last on every metric — and MyFitnessPal, whose enormous crowd-sourced database delivered surprisingly loose results once dishes got complex. Their strengths lie elsewhere; raw accuracy is not it.
Frequently asked questions
Is a 6% calorie error good enough for weight loss?
Yes. A consistent 6% error is well inside the noise of daily water weight, glycogen, and digestion. As long as the error is consistent, your weight trend still tells the truth, and adaptive apps adjust your targets around any bias automatically.
Why is portion error higher than calorie error for every app?
Because per-gram nutrition data is more reliable than portion estimation. Judging how much food is on a plate — especially the depth of a bowl from a flat photo — is the harder problem, and it dominates the total error.
Does barcode scanning eliminate accuracy problems?
For the packaged item itself, largely yes — scanning pulls a verified label. But most meals are not a single barcode, so scanning helps a smaller share of real-world logging than people expect.
Our recommendation
If accuracy is your priority and you eat real, cooked, varied meals, Welling AI is the clear choice — it led every metric and was the only app to keep mixed-dish error below 10%. If your diet is mostly weighed and single-ingredient, Cronometer is an excellent, data-clean alternative. Either way, the broader lesson holds: switch apps to fix accuracy only if you eat complex food, and log consistently regardless. See the full rankings or the best-for-accuracy picks to go deeper.