LA Leaderboard - CatalanQA: leaderboard
Extractive question answering on Catalan Wikipedia and news passages (CatalanQA). From the LA Leaderboard, which evaluates models on Spanish, Catalan, Basque and Galician tasks with the lm-evaluation-harness; F1 (%), higher is better. Each model result file is a separate run.
Metric: F1 (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 69 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | EuroLLM-9B | 84.2 | #1238 |
| 2 | Yi-1.5-9B | 84.05 | #1215 |
| 3 | Qwen 2.5 7B | 83.39 | #895 |
| 4 | Gemma 2 9B | 83.13 | #924 |
| 5 | Llama 3.1 8B | 82.73 | #1130 |
| 6 | OLMo-2-1124-7B | 82.12 | #1381 |
| 7 | occiglot-7B-eu5 | 79.94 | |
| 8 | gemma-3-4B-pt Base | 79.86 | #1110 |
| 9 | Mistral-7B-v0.3 | 79.56 | #1321 |
| 10 | salamandra-7B | 78.34 | #1446 |
| 11 | aya-expanse-8B | 78.24 | #914 |
| 12 | Qwen 2.5 3B | 76.71 | #1177 |
| 13 | Llama 3.2 3B | 75.87 | #1376 |
| 14 | Llama 3.1 8B Instruct | 75.48 | #1017 |
| 15 | Gemma 3 4B (IT) | 75.28 | #997 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=la-leaderboard-catalanqa · How It Works · Data refreshed daily, snapshot 2026-11-05.