Machine Translation on WMT En->Zh (BLEURT-20, R1-judge)
67.21BLEURT-20SeedX-PPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SeedX-PPOModel Category=Translation-Specialized2026.02 | 67.21 | 87.98 | |
| SSR-X-Zero-7BModel Category=Translation-Specialized2026.02 | 66.12 | 83.78 | |
| Gemini-2.5-ProModel Category=General LLM2026.02 | 66 | 91.31 | |
| DeepSeek-R1-0528Model Category=General LLM2026.02 | 64.87 | 89.24 | |
| Qwen2.5-7B + GRPOModel Category=Translation-Specialized, Backbone=Qwen2.5-7B-SFT, Algorithm=GRPO2026.02 | 64.8 | 87.8 | |
| Qwen2.5-7B + GRPO w/ CLAModel Category=Translation-Specialized, Backbone=Qwen2.5-7B-SFT, Algorithm=GRPO with Cross-Lingual Augmentation2026.02 | 63.91 | 88.29 | |
| TowerInstruct-13BModel Category=Translation-Specialized2026.02 | 62.92 | 77.63 | |
| Qwen2.5-7B-SFTModel Category=Translation-Specialized2026.02 | 59.99 | 76.98 | |
| Qwen2.5-7B-InstructModel Category=General LLM2026.02 | 59.92 | 80.51 | |
| Qwen3-8BModel Category=General LLM2026.02 | 57.58 | 84.15 |