VNTL Leaderboard — leaderboard
Japanese visual-novel translation benchmark where LLMs and translation systems render 256 samples into English while preserving semantic similarity.
Metric: Accuracy (%). Source: huggingface.co. Status: saturation imminent. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4o (2024-05-13) | 75.16 |
| 2 | GPT-4o (2024-08-06) | 74.97 |
| 3 | Claude 3 Opus | 74.59 |
| 4 | Claude 3.5 Sonnet (20240620) | 74.4 |
| 5 | DeepSeek V3 Chat | 74.24 |
| 6 | Claude 3.5 Sonnet (20241022) | 72.8 |
| 7 | GPT-4o Mini (2024-07-18) | 72.23 |
| 8 | Grok 2 (1212) | 71.6 |
| 9 | Grok Beta | 71.27 |
| 10 | DeepSeek V2.5 | 71.14 |
| 11 | Qwen 2 72B Instruct | 70.2 |
| 12 | GPT-3.5 Turbo (1106) | 69.98 |
| 13 | Llama 3.1 70B Instruct | 69.79 |
| 14 | Llama 3.1 405B Instruct | 69.46 |
| 15 | GPT-4 (0613) | 69.28 |
Interactive version: theaggregate.ai/benchmark?slug=vntl-leaderboard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.