The 2026 AI Food Tracker Benchmark
15,000 meal photos. 10 cuisines. Three difficulty tiers. Every app submitted blind, three times, with the median taken.
15,000
Meal photos in the test set, captured under fixed lighting on an iPhone 15 Pro.
±0.1g
Every portion lab-weighed to a tenth of a gram before photography.
10 cuisines
Italian, Japanese, Mexican, Indian, Levantine, Thai, West African, US, French, Chinese.
Triple-submit
Each photo submitted three times per app from clean accounts; median result scored.
The scoring formula
Five weighted dimensions sum to a single composite score out of 10.
Difficulty tiers
Single items
A pear. A bowl of rice. A grilled chicken breast. Tests baseline recognition.
2–3 overlapping items
Composed plates where segmentation matters: salad with protein, rice with two sides.
Mixed dishes with sauces
Curry, ragù, gumbo, biryani. Where most trackers collapse — and where Welling separates.
Final results
Every tracker, every weighted dimension, single ranked table.
| # | Tracker | Score | ID Rate | Portion Error | Speed | Foods |
|---|---|---|---|---|---|---|
| 1 | Welling Editor's pick | 9.7 / 10 | 95.6% | ±1.2% | 2.6s | 3,400+ |
| 2 | MyFitnessPal | 7.8 / 10 | 72.4% | ±17% | 8.7s | 2,800+ |
| 3 | Lose It! | 7.5 / 10 | 67.3% | ±23% | 11.6s | 1,900+ |
| 4 | MacroFactor | 7.4 / 10 | 66.2% | ±21% | 10.2s | 1,200+ |
| 5 | Cronometer | 7.3 / 10 | 64.8% | ±22% | 12.4s | 950+ |
| 6 | Cal AI | 7.1 / 10 | 63.5% | ±25% | 9.4s | 1,500+ |
| 7 | SnapCalorie | 7.0 / 10 | 61.7% | ±27% | 5.9s | 2,300+ |
| 8 | Fitia | 6.9 / 10 | 59.3% | ±29% | 8.1s | 2,100+ |
| 9 | Foodvisor | 6.8 / 10 | 57.6% | ±32% | 7.8s | 2,600+ |
| 10 | BitePal | 6.5 / 10 | 55.1% | ±35% | 14.2s | 900+ |