Dicta-LM 3.0: Advancing The Frontier of Hebrew Sovereign LLMs
Shaltiel Shmidman, Avi Shmidman, Amir DN Cohen Moshe Koppel · 2026
Read Dicta-LM 3.0: Advancing The Frontier of Hebrew Sovereign LLMs by Shaltiel Shmidman, Avi Shmidman, Amir DN Cohen Moshe Koppel (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02104.txt.
Description
Shaltiel Shmidman and colleagues present Dicta-LM three point zero, an open-weight set trained on Hebrew and English. Open-weight models abound, yet sovereign models stay scarce. Sizes run 24B on Mistral-Small three point one, 12B on NVIDIA Nemotron Nano V2, and one point seven B on Qwen3, each in base and chat forms, with 65k-token context. To test the models strictly, the team adds a new Hebrew test suite spanning Translation, Summarization, Winograd, and Diacritization. The work sketches a path for other tongues.
Catalog record
| CREATOR | Shaltiel Shmidman, Avi Shmidman, Amir DN Cohen Moshe Koppel |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36470 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.02104