AI benchmarks and leaderboards are different things
Learn the difference between a benchmark, an evaluation run and a leaderboard, and use each to answer the right question.
OpenCompass Hub
HellaSwag
HellaSwag is a challenge dataset for evaluating commonsense natural language inference, which is specially hard for state-of-the-art models, though its questions are trivial for humans (>95% accuracy). It consists of 70k multiple choice questions, each with a scenario and four possible endings, which requires to select the most reasonable ending. These questions come from two domains:activitynet and wikihow, involving video and text scenarios respectively. The correct answers of these questions are the real sentences for the next event, while the incorrect answers are adversarially generated and human verified, so as to fool machines but not humans.
Evaluation context and evidence
- Source
- OpenCompass Hub
- Publisher
- Not recorded
- Benchmark release
- 2019-05-19
- Collected
- 2026-08-18
- Modality
- Not recorded
These are source-reported observations, not independent reruns by this site. Check the task version, evaluation protocol, unit and reporting date before comparing models. Results from different setups are not a universal ranking.
Original sources
Reported results 0
This source record has no numeric results in our collection. That does not mean a score of zero or that no public results exist. Consult the original resources above.