Multitask Language Understanding on MMLU Spanish (test)
77AccuracyTrinity Large
Evaluation results
| Method | Links | |
|---|---|---|
| Trinity LargeArchitecture=MoE, Protocol=Zero-shot2026.02 | 77 | |
| Qwen3Parameters=8B, Protocol=Zero-shot2026.02 | 71 | |
| Qwen3Parameters=4B, Protocol=Zero-shot2026.02 | 67 | |
| DatologyParameters=8B, Protocol=Zero-shot2026.02 | 55 | |
| Llama-3.1Parameters=8B, Protocol=Zero-shot2026.02 | 55 | |
| Granite-4.0Parameters=Micro, Protocol=Zero-shot2026.02 | 54 | |
| SmolLM3Parameters=3B, Protocol=Zero-shot2026.02 | 53 | |
| LFM2.5Parameters=1.2B, Protocol=Zero-shot2026.02 | 50 | |
| Llama-3.2Parameters=3B, Protocol=Zero-shot2026.02 | 47 | |
| DatologyParameters=3B, Protocol=Zero-shot2026.02 | 46 | |
| MLFiltering Strategy=Multilingual (ML) classifier, Model size=1B2026.04 | 30.5 | |
| Llama-3.2Parameters=1B, Protocol=Zero-shot2026.02 | 30 | |
| HQFiltering Strategy=Monolingual high-quality (HQ), Model size=1B2026.04 | 29.85 | |
| No filteringFiltering Strategy=None, Model size=1B2026.04 | 28.14 |