This is the model card for EuroLLM1.5-9B-Base.
We compare EuroLLM 1.5 9B against previous EuroLLM base models (EuroLLM 9B and EuroLLM 22B, both from the 2512 release), as well as two European open base models of comparable size, Apertus 1.5 8B and OpenEuroLLM Prelude 9B. Our evaluation covers English and multilingual benchmarks assessing mathematical reasoning, general reasoning, coding, knowledge, and reading comprehension.
Note: The results below are preliminary and subject to further verification. Final results and detailed evaluation configurations will be released soon to ensure reproducibility.
| EuroLLM1.5-9B-Base | EuroLLM-9B-2512 | EuroLLM-22B-2512 | Apertus-v1.5-8B-base | OpenEuroLLM Prelude-base 9B | |
|---|---|---|---|---|---|
| English | |||||
| Math | |||||
| GSM8K | 64.3 | 45.6 | 56.5 | 60.0 | 79.7 |
| Reasoning | |||||
| ARC-C | 62.7 | 58.9 | 64.1 | 64.7 | 65.1 |
| HellaSwag | 76.9 | 75.6 | 77.5 | 76.9 | 78.0 |
| PIQA | 81.5 | 80.5 | 81.9 | 80.7 | 82.6 |
| WinoGrande | 73.7 | 70.7 | 75.7 | 71.8 | 75.2 |
| Coding | |||||
| HumanEval | 41.5 | 28.1 | 29.3 | 31.7 | 60.4 |
| MBPP | 51.4 | 37.4 | 44.4 | 50.0 | 55.0 |
| Knowledge | |||||
| MMLU | 65.8 | 52.3 | 62.7 | 63.7 | 68.0 |
| Multilingual | |||||
| Math | |||||
| MGSM | 64.2 | 37.0 | 51.4 | 48.3 | 57.0 |
| Reading Comprehension | |||||
| Belebele | 81.7 | 69.6 | 80.8 | 78.3 | 82.9 |
| Knowledge | |||||
| Global-MMLU | 58.2 | 47.3 | 55.7 | 55.5 | 58.4 |
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "utter-project/EuroLLM1.5-9B-Base"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype="auto")
text = "English: My name is EuroLLM. Portuguese:"
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=20)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
EuroLLM1.5-9B-Base is a base model: it has not been instruction-tuned or aligned to human preferences, so the model may generate problematic outputs (e.g., hallucinations, harmful content, or false statements).
If you use our work, please cite the following until the EuroLLM1.5 technical report is released:
@misc{ramos2026eurollm22btechnicalreport,
title={EuroLLM-22B: Technical Report},
author={Miguel Moura Ramos and Duarte M. Alves and Hippolyte Gisserot-Boukhlef and João Alves and Pedro Henrique Martins and Patrick Fernandes and José Pombal and Nuno M. Guerreiro and Ricardo Rei and Nicolas Boizard and Amin Farajian and Mateusz Klimaszewski and José G. C. de Souza and Barry Haddow and François Yvon and Pierre Colombo and Alexandra Birch and André F. T. Martins},
year={2026},
eprint={2602.05879},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2602.05879},
}
model-00001-of-00004.safetensors 5.0 GBmodel-00003-of-00004.safetensors 4.9 GBmodel-00002-of-00004.safetensors 4.9 GBmodel-00004-of-00004.safetensors 3.5 GB