기사 목록으로 돌아가기
Models10 min
Measuring DeepSeek's Japanese: JMMLU, JCommonSenseQA, AI-Oh Benchmarks
Misaki HayashiAI Research Analyst2026-04-2010 min
DeepSeekJapanese NLPBenchmarksJMMLUEvaluation
이 글은 일본어로 작성되어 있습니다. 한국어 요약은 아래와 같습니다:
Measuring DeepSeek's Japanese: JMMLU, JCommonSenseQA, AI-Oh Benchmarks—Internal benchmark of DeepSeek V3.2/R1 vs Qwen2.5-72B, Claude 3.5 Sonnet and GPT-4o-mini on JMMLU, JCommonSenseQA and AI-Oh, with reproducible prompt formats and conditions.