컬렉션 살펴보기

AI 벤치마크

도구 14개 · 1개 그룹

AI 모델 벤치마크14

AI 벤치마크 고르는 법

AI benchmarks are platforms and test suites that score and rank large models on the same questions and rules, giving you an objective reference when choosing a model instead of relying on vendor claims.

What's on this page

  • Human-vote leaderboards: LMArena has users compare two anonymous models' answers and vote, ranking models by the results — closer to real-world preference.
  • Academic benchmarks: MMLU (massive multitask language understanding), C-Eval and CMMLU (Chinese), MMBench (multimodal) and others test knowledge and reasoning with fixed question sets.
  • Evaluation platforms: the Open LLM Leaderboard from Hugging Face, OpenCompass from Shanghai AI Laboratory, HELM, FlagEval, SuperCLUE and others combine several benchmarks into an overall view.

Reading leaderboards

  • Different boards measure different things — knowledge, coding, maths, Chinese, long context — so focus on the ones that match your use.
  • Public question sets may have been seen during training, inflating scores; human-vote boards are closer to real experience but are also influenced by answer style.
  • Treat leaderboards as a guide; the most reliable method is to test two or three candidate models on your own real tasks.
AI로 할 수 있는 일을 계속 살펴보세요전체 도구 보기