Advertisement
Compare LLMs optimized for Japanese language. Learn about domestic data processing, Japanese quality, and enterprise compliance advantages.
Optimized for high-quality Japanese language output
Data processed on domestic servers for compliance
Aligned with Japanese enterprise regulations
Strong at keigo (honorifics) and business documents
| Model Name | Developer | Parameters | Japanese Benchmark | License | Best For |
|---|---|---|---|---|---|
| tsuzumi | NTT | 7B | ★★★★★ | Commercial (contract) | Enterprise Japanese processing, on-premise deployment |
| ELYZA LLM | ELYZA (KDDI subsidiary) | 70B | ★★★★☆ | Research / Commercial | Telecom and finance Japanese processing |
| CyberAgentLM3 | CyberAgent | 68B | ★★★★☆ | Apache 2.0 | Ad copy and marketing content |
| PLaMo | Preferred Networks | 100B | ★★★★☆ | Commercial (contract) | Scientific computing and academic research |
| Swallow | Tokyo Tech / AIST | 70B | ★★★★☆ | Llama 2 License | Academic research and Japanese NLP tasks |
| KARAKURI LM | KARAKURI | 70B | ★★★☆☆ | Commercial (contract) | Customer support and FAQ |
NTT
Lightweight high-performance LLM by NTT. 7B parameters optimized for Japanese. Supports edge device inference.
ELYZA (KDDI subsidiary)
Llama 2-based with Japanese fine-tuning. Proven in KDDI's telecom and finance sectors.
CyberAgent
Japanese LLM strong in advertising and marketing. 68B parameters optimized for creative text generation.
Preferred Networks
Built by deep learning pioneer PFN. Strong in math and science tasks, optimized for Japanese academic use.
Tokyo Tech / AIST
Llama 2-based continual pre-training model for Japanese. Widely used in research with high Japanese benchmark scores.
KARAKURI
Customer support specialized Japanese LLM. Optimized for FAQ response and chatbot construction.
Comparing domestic Japanese LLMs with global models (GPT-4o, Claude, Gemini) on Japanese tasks
OpenAI
Anthropic