Machine Translation on WMT 24++ (Selected Language Pairs)
72.86AR ScoreGPT-5-mini
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5-miniModel Class=Baselines2026.04 | 72.86 | 91.71 | 86.93 | 84.3 | 86.34 | 82.15 | 82.56 | 80.29 | 79.49 | 77.76 | 82.44 | |
| Qwen3-235B-A22BModel Class=Baselines2026.04 | 69.72 | 90.42 | 85.53 | 81.93 | 84.59 | 78.54 | 79.76 | 77.32 | 73.92 | 75.88 | 79.76 | |
| Qwen3-14B + EA-RLVRBackbone=Qwen3-14B, Training Protocol=EA-RLVR, Data Scale=7k standard2026.04 | 68.3 | 89.79 | 84.28 | 80.78 | 83.56 | 77.61 | 77.49 | 76.3 | 70.16 | 75.84 | 78.41 | |
| Qwen3-14B + EA-RLVR (full data)Backbone=Qwen3-14B, Training Protocol=EA-RLVR, Data Scale=full2026.04 | 68 | 90.11 | 85.06 | 81.27 | 84.1 | 78.14 | 77.57 | 75.97 | 70.25 | 75.99 | 78.65 | |
| Qwen3-14BBackbone=Qwen3-14B, Training Protocol=Base2026.04 | 66.37 | 89.29 | 83.79 | 80.15 | 81.91 | 76.02 | 75.73 | 74.87 | 67.4 | 75.1 | 77.06 | |
| Qwen3-14B + SFTBackbone=Qwen3-14B, Training Protocol=SFT, Data Scale=7k standard2026.04 | 66.2 | 89.12 | 84.1 | 80.09 | 82.37 | 76.38 | 75.35 | 75.23 | 67.7 | 75.15 | 77.17 | |
| Qwen3-8B + EA-RLVR (full data)Backbone=Qwen3-8B, Training Protocol=EA-RLVR, Data Scale=full2026.04 | 65.29 | 89 | 83.48 | 79.82 | 82.17 | 74.52 | 73.62 | 73.88 | 65.58 | 74.06 | 76.14 | |
| Qwen3-14B + SFT (full data)Backbone=Qwen3-14B, Training Protocol=SFT, Data Scale=full2026.04 | 65.27 | 89.11 | 84.07 | 79.87 | 82.47 | 76.26 | 75.5 | 74.92 | 67.82 | 75.18 | 77.05 | |
| Qwen3-8B + EA-RLVRBackbone=Qwen3-8B, Training Protocol=EA-RLVR, Data Scale=7k standard2026.04 | 64.65 | 88.85 | 83.29 | 79.48 | 81.16 | 73.6 | 74.04 | 72.92 | 64.91 | 74.2 | 75.71 | |
| Qwen3-8B + SFTBackbone=Qwen3-8B, Training Protocol=SFT, Data Scale=7k standard2026.04 | 62.73 | 88.29 | 82.91 | 78.52 | 80.23 | 72.99 | 71.73 | 72.2 | 62.47 | 73.67 | 74.57 | |
| Qwen3-8B + SFT (full data)Backbone=Qwen3-8B, Training Protocol=SFT, Data Scale=full2026.04 | 62.51 | 87.78 | 82.25 | 78.79 | 80.48 | 72.32 | 71.37 | 71.57 | 61.83 | 73.31 | 74.22 | |
| Qwen3-8BBackbone=Qwen3-8B, Training Protocol=Base2026.04 | 61.88 | 88.06 | 82.36 | 78.33 | 81.09 | 72.86 | 71.97 | 72.43 | 62.48 | 73.11 | 74.46 | |
| DeepTrans-7BModel Class=Baselines2026.04 | 56.51 | 84.69 | 81.45 | 76.31 | 74.86 | 70.24 | 62.94 | 65.8 | 48.78 | 71.53 | 69.31 | |
| Marco-o1Model Class=Baselines2026.04 | 53.5 | 84.09 | 80.23 | 75.38 | 74.84 | 67.04 | 67.04 | 67.04 | 48.15 | 70.47 | 68.78 |