Machine Translation on 1,700-passage translation dataset high-drift instances
87.9Valid Translation RateGPT-5.5
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-5.5High-drift n=2242026.05 | 87.9 | 8 | 4 | 2.1 | 12.1 | |
| Gemini 3.1 ProHigh-drift n=2572026.05 | 79 | 14 | 7 | 4.5 | 21 | |
| Claude Sonnet 4.6High-drift n=3022026.05 | 77.5 | 13.9 | 8.6 | 5.9 | 22.5 | |
| Grok 4.3High-drift n=4202026.05 | 55 | 17.4 | 27.6 | 23.6 | 45 |