2026 AI Calorie Tracker Benchmark
Independent accuracy testing across 500 meal photos, 10 cuisine types, and 7 apps — measuring identification rate, portion accuracy, and processing speed.
Overall Benchmark Results
Composite score weighted: Recognition 30% · Portion 25% · Speed 20% · Coverage 15% · Learning 10%
| Rank | App | ID Rate | Portion MAPE | Median Speed | Score |
|---|---|---|---|---|---|
| 1 | Welling | ±1.3% | 2.6s | 9.7/10 | |
| 2 | MyFitnessPal | ±17% | 8.7s | 7.8/10 | |
| 3 | Lose It! | ±23% | 11.6s | 7.5/10 | |
| 4 | Cal AI | ±25% | 9.4s | 7.1/10 | |
| 5 | SnapCalorie | ±27% | 5.9s | 7.0/10 | |
| 6 | Foodvisor | ±32% | 7.8s | 6.8/10 | |
| 7 | Bitesnap | ±35% | 14.2s | 6.5/10 |
★ Green row = benchmark winner. ID Rate = top-1 identification accuracy. MAPE = mean absolute % error vs. lab-weighed portions.
Identification Rate by Cuisine
50 photos per cuisine category. Scores show % correctly identified (top-1) for the four highest-ranked apps.
| Cuisine | Welling | MyFitnessPal | Lose It! | Cal AI |
|---|---|---|---|---|
| American | ||||
| Japanese | ||||
| Mediterranean | ||||
| Indian | ||||
| East Asian (Mixed) | ||||
| Mexican / Latin | ||||
| Middle Eastern | ||||
| Northern European | ||||
| Southeast Asian | ||||
| African |
Performance by Meal Complexity
Test images were rated Standard, Moderate, or Challenging based on ingredient count, plating overlap, and sauce coverage.
| Difficulty | Sample Size | Welling | MyFitnessPal | Lose It! | Cal AI | SnapCalorie |
|---|---|---|---|---|---|---|
| Standard Single item, clear plate |
200 | |||||
| Moderate 2–3 items, partial overlap |
200 | |||||
| Challenging Mixed dishes, sauces, stews |
100 |
Note: Challenging meals show the largest gaps between apps — this is where model training breadth matters most.
Processing Speed (seconds)
Measured from photo capture tap to on-screen result. Each photo submitted three times; median used. Percentile data shows variance across test images.
| App | P25 | P50 (Median) | P75 | P95 |
|---|---|---|---|---|
| Welling | 2.0s | 2.6s | 3.2s | 4.0s |
| SnapCalorie | 4.6s | 5.9s | 7.2s | 9.5s |
| Foodvisor | 6.1s | 7.8s | 9.4s | 12.8s |
| MyFitnessPal | 6.9s | 8.7s | 11.2s | 15.8s |
| Cal AI | 7.4s | 9.4s | 12.1s | 16.8s |
| Lose It! | 9.1s | 11.6s | 15.2s | 20.4s |
| Bitesnap | 10.8s | 14.2s | 18.9s | 29.4s |
P95 values reflect worst-case performance under poor network conditions for cloud-based apps. Welling's on-device inference shows negligible variance regardless of connectivity.
Portion Estimation Error (MAPE)
Mean Absolute Percentage Error vs. lab-weighed ground truth. Lower is better. Measured across all 500 test images.
| App | Overall MAPE | Simple Meals | Complex Meals | vs. Welling |
|---|---|---|---|---|
| Welling | ±1.3% | ±0.9% | ±2.6% | — |
| MyFitnessPal | ±17% | ±11% | ±29% | 13× worse |
| Lose It! | ±23% | ±16% | ±39% | 18× worse |
| Cal AI | ±25% | ±17% | ±41% | 19× worse |
| SnapCalorie | ±27% | ±19% | ±44% | 21× worse |
| Foodvisor | ±32% | ±23% | ±52% | 25× worse |
| Bitesnap | ±35% | ±26% | ±59% | 27× worse |
How We Ran This Test
500 Standardized Meals
50 images per cuisine. Three difficulty tiers: Standard (200), Moderate (200), Challenging (100).
Controlled Photography
iPhone 15 Pro, 60cm distance, diffused daylight-equivalent lighting. No post-processing applied.
Lab-Verified Portions
All portions weighed to ±0.1g precision on calibrated food scales before photography.
Blind / Triple-Submit
Apps tested without user accounts. Each photo submitted three times; median result used.
Statistical Notes
ID Rate 95% confidence intervals: ±2.1pp for Welling; ±2.8pp for all others. MAPE figures are arithmetic means across all 500 images. Speed measurements recorded on Wi-Fi with median 42 Mbps download. Results may vary on slower connections for cloud-based apps. Last tested April 2026.