Japanese G2P (Parse Mode): leaderboard
Japanese grapheme-to-phoneme conversion (2026): LLMs read 3,000 manually annotated JVS corpus sentences into kana, the pronunciation input for text-to-speech. In parse mode the LLM segments each sentence and gives word readings, and rule-based post-processing applies particle and long-vowel rules; the score is kana character error rate in percent, lower is better.
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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 0.52 | #102 |
| 2 | Gemini 3.1 Pro (Preview) (Low) | 0.62 | #94 (Gemini 3.1 Pro (Preview)) |
| 3 | Claude Sonnet 4.6 | 0.82 | #161 |
| 4 | Gemini 3 Flash (Low) | 0.94 | #164 (Gemini 3 Flash) |
| 5 | Gemini 2.5 Flash (Non-reasoning) | 1.08 | #453 (Gemini 2.5 Flash) |
| 6 | Kimi K2.5 (Non-reasoning) | 1.35 | #256 (Kimi K2.5) |
| 7 | GLM-5 (Non-reasoning) | 1.44 | #260 (GLM-5) |
| 8 | GPT-5.2 (Non-reasoning) | 1.64 | #176 (GPT-5.2) |
| 9 | Qwen 3.5 397B A17B (Non-reasoning) | 1.96 | #246 (Qwen 3.5 397B A17B) |
| 10 | GPT-OSS-120B (Low) | 3.03 | #700 (GPT-OSS-120B) |
| 11 | GLM-4.7 (Non-reasoning) | 3.7 | #366 (GLM-4.7) |
| 12 | Gemma 3 27B | 5.75 | #1631 |
| 13 | Qwen 3.5 35B A3B (Non-reasoning) | 6.12 | #482 (Qwen 3.5 35B A3B) |
| 14 | Qwen 3.5 27B (Non-reasoning) | 6.27 | #359 (Qwen 3.5 27B) |
| 15 | Llama 3.3 70B Instruct | 6.58 | #1357 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=japanese-g2p-parse-mode · How It Works · Data refreshed daily, snapshot 2026-11-13.