The repository maintains a curated list of pre-trained models specifically adapted for the legal domain. These models vary by language (e.g., Hebrew, English, Italian, French, Chinese), size, and architecture (e.g., BERT, RoBERTa, Longformer, LLaMA).
Key model categories include:
- Domain-specific BERTs: Such as
LEGAL-BERT, JuriBERT, and Italian-LEGAL-BERT. - Large Language Models (LLMs): Such as
LEGAL-GPT and Legal LLaMA. - Multilingual/Multijurisdictional: Such as
MultiLegalPile Models and FairLex Domain Models. - Task-specific: Such as
LegalCore Models for event coreference resolution.
| Model | Links | Language | Size |
|---|---|---|---|
| Legal-HeBERT (Chriqui et al., 2022) | [📄](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4147127) [🤗](https://huggingface.co/avichr/Legal-heBERT) [💻](https://github.com/avichaychriqui/Legal-HeBERT) | 🇮🇱 | 110M |
| PoL-BERT-Large (Henderson et al., 2022) | [📄](https://arxiv.org/abs/2207.00220) [🤗](https://huggingface.co/pile-of-law/legalbert-large-1.7M-1) [💻](https://github.com/Breakend/PileOfLaw) | 🇬🇧 | 336M |
| Italian-LEGAL-BERT (Licari and Comande, 2022) | [📄](https://ceur-ws.org/Vol-3256/km4law3.pdf) [🤗](https://huggingface.co/dlicari/Italian-Legal-BERT)| 🇮🇹 | 110M |
| JuriBERT (Douka et al., 2021) | [📄](https://arxiv.org/abs/2110.01485) [💾](http://master2-bigdata.polytechnique.fr/resources#juribert) | 🇫🇷 | {6M, 15M, 42M, 110M} |
| Custom-LEGAL-BERT (Zheng et al., 2021) | [📄](https://arxiv.org/abs/2104.08671) [🤗](https://huggingface.co/zlucia/custom-legalbert) [💻](https://github.com/reglab/casehold) | 🇬🇧 | 110M |
| LEGAL-BERT (Chalkidis et al., 2020) | [📄](https://arxiv.org/abs/2010.02559) [🤗](https://huggingface.co/nlpaueb/legal-bert-base-uncased) | 🇬🇧 | {35M, 110M} |
| LEGAL-GPT-{1,2} (Borchmann et al., 2020) | [📄](https://arxiv.org/abs/1911.03911) [💻](https://github.com/applicaai/contract-discovery) | 🇬🇧 | {117M, 1.5B} |
| MultiLegalPile Models (2024-2025) | [📄](https://reglab.stanford.edu/publications/multilegalpile/) [🤗](https://huggingface.co/collections/joelniklaus/multilegalpile-datasets-6535db705f5e918bdc17ecc7) | 🌍 | RoBERTa (multilingual + 24 monolingual), Longformer |
| Legal-BERT Fine-tuned (2024) | [📄](https://towardsai.net/p/artificial-intelligence/fine-tuning-legal-bert-llms-for-automated-legal-text-classification) | 🇬🇧 | Domain-adapted classification models |
| LegalCore Models (2025) | [📄](https://aclanthology.org/2025.findings-acl.1284.pdf) | 🌍 | Event coreference resolution for legal texts |
| Legal LLaMA (2025) | [📄](https://arxiv.org/abs/2305.15062) | 🇨🇳 | Chinese legal domain adaptations |
| FairLex Domain Models (2024-2025) | [🤗](https://huggingface.co/collections/coastalcph/legal-nlp-683ffd107b0dc24048c449ae) | 🌍 | Domain-specific BERT models for 4 jurisdictions |
