TruthfulQA Benchmark
Explore score history ↗817 questions (arXiv:2607.28801, Table 1), the full benchmark. Measures how models mimic human falsehoods (arXiv:2109.07958). The study notes the difficulties of consistent human judgment (arXiv:2607.28801, section 2). The paper reports 18.7 / 21.0 over TruthfulQA MC1+MC2 combined (n=1634), not the standard single TruthfulQA metric, so no plain TruthfulQA rows are filed from it; the combined scores are filed under benchmark_id truthfulqa with instrument truthfulqa_mc1_mc2.
- Released
- 2021-09-08
- Evaluates
- factuality
- Openness
- unknown
- Importer
- Model reports
TruthfulQA paper
TruthfulQA results and reported scores
Reported benchmark scores · Model reports
Reported scores: 2
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.
Which sources cite TruthfulQA?
- Llama-3.2-1B and SmolLM-1.7B-Instruct Aleph Alpha Research, 2026-07-30