LLM-as-a-judge evaluation on 100 Romanian synthetic prompts (test)
4.71FluencyDeepSeek (Instruct)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek (Instruct)2026.01 | 4.71 | 4.66 | 4.68 | |
| LLaMA-3 8B InstructParameters=8B2026.01 | 4.7 | 4.72 | 4.71 | |
| GPT-4.1-mini2026.01 | 4.63 | 4.77 | 4.7 | |
| EuroLLM 9B InstructParameters=9B2026.01 | 4.58 | 4.66 | 4.62 | |
| Gemma 2 9B InstructParameters=9B2026.01 | 4.46 | 4.58 | 4.52 | |
| TF3 TransformerParameters=51.65M2026.01 | 4.28 | 4.09 | 4.19 | |
| TF3 Distilled StudentParameters=26.45M2026.01 | 3.96 | 4.03 | 4 |