Meta
Llama 2 70B
63.52%Public score reported by Meta in the Llama 2 technical report (Table 10, MC2 format). Methodology and prompt settings may differ from Data Canon future evaluations.
Source ReportedBenchmark
TruthfulQA
A benchmark measuring whether language models generate truthful answers to 817 questions that probe common misconceptions. Higher scores indicate greater truthfulness and resistance to confidently stating falsehoods.
What it measures
Percentage of truthful answers using MC2 multi-correct format across health, law, finance, science, and common misconceptions
Safety areas: Truthfulness
Score direction: higher_is_better
Public model scores
Meta
Llama 2 13B
62.18%Public score reported by Meta in the Llama 2 technical report (Table 10, MC2 format). Methodology and prompt settings may differ from Data Canon future evaluations.
Source ReportedMeta
Llama 2 7B
57.04%Public score reported by Meta in the Llama 2 technical report (Table 10, MC2 format). Methodology and prompt settings may differ from Data Canon future evaluations.
Source ReportedPrimary source
TruthfulQA: Measuring How Models Mimic Human FalsehoodsStatus: Verified. Last verified: 2026-06-13.