FoodVitals Benchmark: 1,158 Items for Testing LLM Food Knowledge

The FoodVitals Benchmark Dataset contains 1,158 verified items designed to evaluate the foundational food knowledge of Large Language Models. Created by Nilufar Khodzhaeva and last updated in May 2026, it spans three critical areas: nutritional literacy, nutrient-health relations, and food safety. The data is sourced from authoritative regulatory bodies like the USDA FoodData Central and the EFSA Food Health Claims Knowledge Graph.

Use Cases

• Benchmarking LLM numerical reasoning on nutrient comparisons based on the 'quantitative_recall' and 'ranking' tasks mentioned.
• Evaluating LLM understanding of authorized health claims based on the 'Original_Claim' and 'Target_Population' fields.
• Testing LLM application of food safety guidelines based on the described 'regulatory guidelines for food handling and preparation'.
• Assessing model performance across distinct food knowledge domains (nutrition, health, safety) as described in the three-aspect structure.

Strengths

• Contains 1,158 verified benchmark items.
• Data is sourced from authoritative regulatory sources (USDA, EFSA).
• Provides 'ground_truth' and 'verification_data' fields for objective evaluation.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Freshness should be verified as the last update date is in the future (2026-05-28).

Provenance

Source

USDA FoodData Central and EFSA Food Health Claims Knowledge Graph

Collection Method

Data compiled and verified from authoritative regulatory sources to create benchmark questions.

Freshness

Last updated 2026-05-28 09:40:15