JP-TL-Bench - Japanese to English: leaderboard
Japanese-to-English half of JP-TL-Bench (Lin and Lensenmayer, Shisa.AI, 2026): 36 Japanese source texts (15 Easy, 21 Hard) translated into English, each output compared pairwise and reference-free against the frozen Base Set v1.0 anchor translations by gemini-2.5-flash at temperature 0; omitted subjects and objects must be recovered from context. Bradley-Terry strength over all Japanese-to-English matches, mapped to a 0-10 LT score; higher is better.
Metric: LT Score (0-10; anchored pairwise Bradley-Terry vs 20 Base Set v1.0 anchors, gemini-2.5-flash judge). Source: arxiv.org. Saturation forecast: Estimated already saturated. 64 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 9.98 | #141 |
| 2 | Gemini 2.5 Pro | 9.94 | #262 |
| 3 | Gemini 3 Pro (Preview) | 9.94 | #111 |
| 4 | Qwen 3 235B A22B 2507 Instruct | 9.93 | #619 |
| 5 | GPT-OSS-120B | 9.92 | #699 |
| 6 | DeepSeek V3.1 Terminus | 9.89 | #442 |
| 7 | Ministral-3-14B-Instruct-2512 | 9.88 | #1595 |
| 8 | Kimi K2 0905 | 9.86 | #555 |
| 9 | Gemini 2.5 Flash | 9.84 | #452 |
| 10 | Qwen 3 30B A3B 2507 Instruct | 9.77 | #1078 |
| 11 | Gemma 3 27B (IT) | 9.66 | #1306 |
| 12 | GPT-4o (2024-08-06) | 9.65 | #731 |
| 13 | Ling-flash-2.0 | 9 | #1352 |
| 14 | ABEJA-Qwen2.5-32B-Japanese-v1.0 | 8.74 | #1304 |
| 15 | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | 8.69 | #1773 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=jp-tl-bench-japanese-to-english · How It Works · Data refreshed daily, snapshot 2026-11-12.