Precision
Say the correct thing, briefly. No inflated benchmarks, no ornamental output.
Research Lab / Tokyo
Sovereign intelligence, engineered in Japan.
The United States built DeepBlue. The United Kingdom founded DeepMind. China shipped DeepSeek. SAMURAI-1 is Japan's answer: a frontier model that treats Japanese and English as native tongues, tuned for precision over spectacle.
Speak with SAMURAI-1
Online
Ask anything in Japanese or English. SAMURAI-1 replies in the language you use.
Public demo environment. Responses are length limited and no conversation is stored.
Model card
SAMURAI-1, the foundation model
Designed for speed and restraint: long context, equal fluency across Japanese and English, and output without ornament.
- Context window
- 131,072tokens
- Languages
- Japanese / EnglishNative grade in both directions
- Throughput
- ~500tokens per second
- Availability
- API and on premiseBuilt for sovereign deployment
Principles
Three tenets
Restraint
Capability is held, not displayed. The model works quietly, only when called.
Sovereignty
A nation and a company should each hold their own intelligence in their own hands.