M-GATE - Round-Trip Translation (Hard Subset): leaderboard
Metric: Mean normalized judge score over 29 target languages (0-1; hard subset). Source: arxiv.org. 82 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (High) | 0.82 |
| 2 | GPT-5.5 (High) | 0.82 |
| 3 | Claude Fable 5 | 0.81 |
| 4 | GPT-5.5 (Medium) | 0.81 |
| 5 | GPT-5.4 (Medium) | 0.8 |
| 6 | GPT-5 | 0.78 |
| 7 | Claude Opus 4.6 (Thinking) | 0.78 |
| 8 | Gemini 3.1 Pro (Preview) | 0.77 |
| 9 | Gemini 3.5 Flash | 0.77 |
| 10 | GPT-5.4 Mini (High) | 0.77 |
| 11 | Gemini 3.5 Flash (High) | 0.77 |
| 12 | Claude Opus 4.5 (Thinking) | 0.77 |
| 13 | GPT-5.4 Mini (Medium) | 0.76 |
| 14 | GPT-5.2 (High) | 0.76 |
| 15 | Gemini 2.5 Pro | 0.75 |
Interactive version: theaggregate.ai/benchmark?slug=m-gate-round-trip-translation-hard-subset · How It Works · Data refreshed daily, snapshot 2026-09-19.