Discourse-level Chinese-English translation on DiscoX
42.62AccuracyGPT-5
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GPT-5Training=-2026.04 | 42.62 | 12.42 | 12.4 | |
| Gemini2.5-ProTraining=-2026.04 | 40.23 | 11.86 | 13.26 | |
| Claude-4-sonnetTraining=-2026.04 | 38.65 | 11.42 | 11.8 | |
| Qwen3-8B + R2-Write-SFT + RLpTraining=SFT + PPO2026.04 | 23.5 | 10.82 | 10.9 | |
| Qwen3-8B + R2-Write-SFT + RLTraining=SFT + PPO2026.04 | 23.16 | 10.91 | 10.95 | |
| Qwen3-8B + R2-Write-SFTTraining=SFT2026.04 | 22.7 | 10.06 | 10.9 | |
| Qwen3-8B + DistillTraining=SFT2026.04 | 21.7 | 9.3 | 10 | |
| Qwen3-8B + Distill + RLTraining=SFT + PPO2026.04 | 20.42 | 9.84 | 10.2 | |
| Qwen3-8B + R2-Write-LastTraining=SFT2026.04 | 20.14 | 9.2 | 9.24 | |
| Qwen3-8B + RLTraining=PPO2026.04 | 19.3 | 9.53 | 9.95 | |
| LongWriter-Zero-32BTraining=SFT + GRPO2026.04 | 19.2 | 9.74 | 9.2 | |
| Qwen3-8BTraining=-2026.04 | 18.8 | 8.74 | 8.8 | |
| Qwen3-4B + R2-Write-SFT + RLpTraining=SFT + PPO2026.04 | 18.64 | 10.22 | 10.28 | |
| Reverse-Engineering-8BTraining=SFT2026.04 | 18.62 | 9.58 | 9.86 | |
| Qwen3-4B + R2-Write-SFTTraining=SFT2026.04 | 18.22 | 8.48 | 9.4 | |
| Qwen3-4B + RLTraining=PPO2026.04 | 15.26 | 7.86 | 8.96 | |
| Writing-RL-7BTraining=SFT + PPO2026.04 | 15.2 | 7.7 | 7.2 | |
| Longwriter-9BTraining=SFT2026.04 | 14.12 | 7.2 | 7.88 | |
| Qwen3-4BTraining=-2026.04 | 12.82 | 7.7 | 8.2 |