The arXiv technical report for LLM-jp-4 introduces a family of Japanese sovereign large language models. According to the abstract, the project places explicit emphasis on transparency, reproducibility, and practical usability, distinguishing it from purely capability-driven releases.

The model family comprises an 8B dense model and a 32B-A3B model. The abstract does not elaborate on the architecture or training details beyond these parameter counts, though the full report likely contains further specifications.

Because this summary is based on a single source—the report's abstract—there is no independent comparison to other models or claims. The stated focus on openness and sovereign development is notable, but verification of these claims would require reading the full technical report or external evaluations.