Benchmark RadarAI BENCHMARK PROFILE
YOMI-Bench
YOMI-Bench evaluates kanji reading and phonological understanding in LLMs for Japanese through four tasks. It is used in a study assessing multilingual and Japanese-specific models.
- Released
- 2026-07-01
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
There is no standalone public comparison path or shared artifact; the benchmark primarily supports the paper's finding that LLMs struggle with kanji reading.
Motivation
We propose YOMI-Bench, a benchmark for evaluating kanji reading and phonological understanding of large language models (LLMs) for Japanese.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.