Japanese G2P (Parse Mode): leaderboard

Metric: Kana character error rate (%; can exceed 100) in parse mode (the LLM segments the sentence and gives each word's reading as JSON, then rule-based post-processing applies particle and long-vowel rules) on the 3,000 manually annotated sentences of the JVS nonpara30 corpus; kana character error rate against the reference readings after removing punctuation and normalizing long vowels; reasoning disabled (non-thinking mode, or low reasoning effort where it cannot be disabled), few-shot prompts; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 37 models tracked.

Top models

#ModelScore
1Claude Opus 4.60.52
2Gemini 3.1 Pro (Preview) (Low)0.62
3Claude Sonnet 4.60.82
4Gemini 2.5 Flash (Non-reasoning)1.08
5Kimi K2.5 (Non-reasoning)1.35
6GLM-5 (Non-reasoning)1.44
7GPT-5.2 (Non-reasoning)1.64
8Qwen 3.5 397B A17B (Non-reasoning)1.96
9GPT-OSS-120B (Low)3.03
10GLM-4.7 (Non-reasoning)3.7
11Gemma 3 27B5.75
12Qwen 3.5 35B A3B (Non-reasoning)6.12
13Qwen 3.5 27B (Non-reasoning)6.27
14Llama 3.3 70B Instruct6.58
15Qwen 3.5 122B A10B (Non-reasoning)7.57

Interactive version: theaggregate.ai/benchmark?slug=japanese-g2p-parse-mode · How It Works · Data refreshed daily, snapshot 2026-09-29.