Multiple-choice question answering on Multiple-choice benchmarks EU
63.26AccuracyQwen3 14B merge-EU
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen3 14B merge-EUBackbone=Qwen3 14B, Variant=merge-EU2026.03 | 63.26 | |
| Llama 3.1 8B joint-EUBackbone=Llama 3.1 8B, Variant=joint-EU2026.03 | 61.75 | |
| ALIA 40B InstructBackbone=ALIA 40B, Variant=Instruct2026.03 | 60.64 | |
| Llama 3.1 8B merge-EUBackbone=Llama 3.1 8B, Variant=merge-EU2026.03 | 58.36 | |
| Qwen3 14B merge-multiBackbone=Qwen3 14B, Variant=merge-multi2026.03 | 56.37 | |
| Qwen3 8B merge-EUBackbone=Qwen3 8B, Variant=merge-EU2026.03 | 55.81 | |
| Qwen3 14B merge-GLBackbone=Qwen3 14B, Variant=merge-GL2026.03 | 53.96 | |
| Qwen3 14B merge-CABackbone=Qwen3 14B, Variant=merge-CA2026.03 | 53.85 | |
| Qwen3 14B merge-ESBackbone=Qwen3 14B, Variant=merge-ES2026.03 | 53.52 | |
| Qwen3 14B InstructBackbone=Qwen3 14B, Variant=Instruct2026.03 | 52.09 | |
| Llama 3.1 8B merge-multiBackbone=Llama 3.1 8B, Variant=merge-multi2026.03 | 51.66 | |
| Llama 3.1 8B InstructBackbone=Llama 3.1 8B, Variant=Instruct2026.03 | 49.29 | |
| Llama 3.1 8B merge-CABackbone=Llama 3.1 8B, Variant=merge-CA2026.03 | 48.26 | |
| Llama 3.1 8B merge-GLBackbone=Llama 3.1 8B, Variant=merge-GL2026.03 | 47.56 | |
| Salamandra 7B InstructBackbone=Salamandra 7B, Variant=Instruct2026.03 | 44.94 | |
| Qwen3 8B merge-multiBackbone=Qwen3 8B, Variant=merge-multi2026.03 | 44.85 | |
| Qwen3 8B InstructBackbone=Qwen3 8B, Variant=Instruct2026.03 | 44.06 | |
| Qwen3 8B merge-ESBackbone=Qwen3 8B, Variant=merge-ES2026.03 | 41.03 | |
| Llama 3.1 8B merge-ESBackbone=Llama 3.1 8B, Variant=merge-ES2026.03 | 40.46 | |
| Qwen3 8B merge-CABackbone=Qwen3 8B, Variant=merge-CA2026.03 | 39.29 | |
| Qwen3 8B merge-GLBackbone=Qwen3 8B, Variant=merge-GL2026.03 | 38.98 | |
| Salamandra 2B InstructBackbone=Salamandra 2B, Variant=Instruct2026.03 | 27.95 |