Accuracy, published

These are the real numbers from Renzy's scanner against an open dataset. No marketing — just MAE, MAPE and RMSE refreshed weekly.

MAPE on kcal

41.9%

Mean absolute percentage error on total calories.

MAE on kcal

Mean absolute error. Lower is better.

MAPE on grams

44%

Portion estimation is the bottleneck.

How it's computed

Each dataset entry is a real-plate photo with weight measured on a kitchen scale and calories reconstructed from USDA. We run the production pipeline exactly like a user scan — no shortcuts, no extra context, no cache.

Open dataset: each photo + ground-truth is documented at /api/admin/accuracy-dataset.
Identical pipeline to production: vision → garnish filter → sauce + brand override → scale anchor → critique → USDA → confirm drink.
Standard metrics: MAPE, MAE, RMSE. No outlier filtering — the worst cases also appear on this page.
Re-run weekly automatically. If a model iteration regresses, you'll see it here before we do.

Error distribution

From the dataset's plates, this is the distribution of percent error on kcal. The tighter the bars cluster on the left, the better.

0–5%

5–10%

10–20%

20–30%

30–50%

50%+

The 5 worst cases

What we got most wrong in this batch. Publishing them forces us to fix them in the next iteration.

Yogur griego natural Hacendado

Truth: 145 kcal · Predicted: 340 kcal

Error: 134.5%

Caña de cerveza 200ml

Truth: 86 kcal · Predicted: 155 kcal

Error: 80.2%

Pasta carbonara 320g

Truth: 560 kcal · Predicted: 836 kcal

Error: 49.3%

1 manzana mediana

Truth: 95 kcal · Predicted: 118 kcal

Error: 24.2%

Tostada con aguacate y huevo poché

Truth: 380 kcal · Predicted: 315 kcal

Error: 17.1%

Last run: 9/14/2026, 6:34:10 AM

Sample size: 8 platos

Model: anthropic/claude-opus-5

Batch: cron-2026-09-14-mu0v5pbz

Want to replicate the benchmark with your own dataset? Email us at hola@renzy.app.