Nejumi 4 - GLP - Translation: leaderboard
Sentence translation in both directions between English and Japanese, on the Asian Language Treebank parallel corpus from the jaster harness. Each direction is scored with the learned COMET wmt22 metric against the reference translation and averaged over the 0-shot and 2-shot runs, and the board is the mean of the two directions on a 0 to 100 scale, higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Estimated already saturated. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 91.65 | #16 |
| 2 | Gemini 3.1 Pro (Preview) | 91.52 | #58 |
| 3 | GPT-5.6 Sol | 91.48 | #19 |
| 4 | GPT-5.5 (2026-04-23) | 91.47 | #22 |
| 5 | Gemini 3 Pro (Preview) | 91.44 | #66 |
| 6 | Gemini 3.5 Flash | 91.44 | #60 |
| 7 | Claude Opus 4.6 | 91.4 | #63 |
| 8 | Claude Opus 4.5 | 91.4 | #83 |
| 9 | Gemini 3.6 Flash | 91.4 | #51 |
| 10 | GPT-5.6 Terra | 91.38 | #41 |
| 11 | Claude Opus 4.7 | 91.37 | #49 |
| 12 | Claude Opus 4 | 91.37 | #163 |
| 13 | Gemini 3 Flash (Preview) | 91.36 | #82 |
| 14 | Claude Opus 4.1 | 91.36 | #140 |
| 15 | Claude Opus 5 | 91.34 | #12 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nejumi-4-glp-translation · How It Works · Data refreshed daily, snapshot 2026-10-20.