
PFN Launches Japan's First From-Scratch Reasoning LLM
Preferred Networks released the beta of PLaMo 3.0 Prime, Japan's first fully from-scratch large language model. It supports advanced 'long thinking' reasoning, inspired by DeepSeek R-1, and approaches performance of Qwen3-235B and gpt-oss-120b. This marks a milestone for domestic LLM development without relying on existing models.







