Lara Mini Base: Language Diversity as a Resource, Not a Trade-off
General-purpose language models are trained mostly on English: over open models we review, at most, non-English pretraining data makes up only a 12% share. Models built for European languages give them far more, but score well below general-purpose models of the same size. A common concern is that, in small models, languages compete for capacity. We present Lara Mini base, a model trained from scratch on 6T tokens in English and 44 other languages, which together cover all 24 official languages of the European Union. As training progresses, the 44 other languages together come to outweigh English. To give these languages enough capacity at low cost, Lara Mini base uses a mixture of experts architecture with 7.2B total and 0.6B active parameters. Among open base models with up to 2B active parameters, it achieves the highest average score across standard English, multilingual, math, and code benchmarks, ahead of models trained on 4 to 6 times as many tokens. Its margin is largest on multilingual benchmarks in the languages it supports and on math benchmarks. It supports contexts of up to 128k tokens. We find that our MoE architecture reduces capacity competition between languages. For a single user on one GPU, without quantization, it generates text at least 1.8 times faster than any model with a higher average score. We release the weights of Lara Mini base under the CC BY-NC-SA 4.0 license. In addition, we will release T-HellaSwag, a higher-quality machine translation of the commonsense reasoning benchmark HellaSwag into 110 language variants.
Authors
- Anna Maria Foti
- Luca Barone
- Giulia Chiriatti
- Vincenzo Di Cicco
- Giulia Boccalini
- Francesco Elefante
- Andrea Blasi (ORCID: https://orcid.org/0000-0003-4791-8794)
- Alessandra Bigi (ORCID: https://orcid.org/0000-0002-1067-6288)
- E. Cappelli
- Paola Cellini
- Andrea Di Paola (ORCID: https://orcid.org/0000-0002-2189-8644)
- Claudia Di Lorenzo
- Francesco Brigante
- Marco Comunità (ORCID: https://orcid.org/0000-0001-8097-9877)
- Emma Gamba (ORCID: https://orcid.org/0000-0003-1662-7804)
- Simona Ferri
- Marta Arena
- Nicola Bertoldi (ORCID: https://orcid.org/0000-0003-4330-5077)
- Michela Di Clemente
- Marta Giani
- Claudia Giannetti
- Alessandro Fusacchia
- Francesco Gasperoni
- Davide Caroselli
- Francesca Carducci
- Valerio Giannini
- Federica Di Antonio (ORCID: https://orcid.org/0009-0007-1552-0737)
- Giorgio Biancini (ORCID: https://orcid.org/0009-0009-6837-0474)
- Cattelan Alessandro
- Andrea De Angelis
- Mauro Cassani
- Tommaso Bonomo
- Eleonora Graziani
- Andrea Ferrari
- Cody Carlson
- Translated
- Anna Cracco
- Kathleen Compares
- Annalisa De Petra
- Helen Rebecca Corfield
- Brenda Patricia Byrne
- Ylenia Di Gese
- Alessandro Falappa
- Daniele Cocozza
- Viviana Fantozzi
- Niccolò Fransoni
- Andrea Castro
- Giulia Ceccacci
- Alan Di Francesco
- Michele Cortellessa
- Francesca Basiricò
- Veronica Carascosi
- Arianna De Pilla
- Antonio Farina
- Francesca Bee
- Sofia Bertini
- Jessica Dickson
- Aurora Cuppone
- Giorgia Campora
- Stefano Florio
- Marta Hontanon Del Olmo
- Davide Gattini
- Mariachiara Fornaro
- Enrico Boscardi
- Dana Barras Franco
- Pierluigi Di Cianni
- Andrea Fernandez Tostado Ro. Diego Andrea
- Stephen William Harris
- Rowena Elizabeth Hogg
- Mara Gramendola
- Joris Demiraj
- Elena Deufemia
- Davide Conti
- Romolo Buondestino
- Filomena Guglielmi
- Daniele Antonini
- Grazia Fiorella Anfuso
- Maria Pia Adinolfi
- L'Meese Hope Greaney
- Rebecca De Giorgi
- Maurizio Davoli
- Federico Cialini
- Melania Annunziata
- Isabelle Andrieu
- Fabio Fioramonti
- Arianna D'Andrea
- Valentina Ciarlariello
- Federica Di Iulio
- Sara Ferri
- Sebastien Bratieres
- Ilaria Fabrizio
- Dario Gubernari
- Antonella Giudici
- Jonathan Hammond
- Giulia Di Pilla
- Emanuela Coghe
- Michele Grandinetti
- Ambra Cascio
- Giulia Brignola
- Dario Cascio
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-12
- DOI
- https://doi.org/10.48550/arxiv.2611.16956
- Primary Topic
- Topic Modeling
- Type
- preprint