たかね (Takane)
An enterprise-oriented domestic large-scale language model developed by Fujitsu. With 13B parameters, it adopts a strategy to provide customization specialized for specific domains such as finance, healthcare, and legal affairs.
파라미터
13B
컨텍스트
8K
라이선스
Proprietary
출시일
2025-11-01
일본어 처리 능력
Model developed by a Japanese company or specialized for Japanese. Highest Japanese understanding and generation capability.
API 가격
이 모델의 API 가격 정보는 현재 공개되지 않았습니다
강점
- ・Enables domain-specific customization
- ・Supports finance, medical, and legal fields
- ・Integrated with Fujitsu's SI business
- ・High security standards
약점
- ・Limited external availability
- ・Not open-source
- ・Few evaluations on general benchmarks
- ・High barrier for small and medium enterprises, geared toward large corporations
활용 사례
- ・AI solutions for the financial domain
- ・Specialized processing in medical/legal fields
- ・Internal AI for large corporations
- ・Safe AI use in regulated industries
심층 분석
Architecture
Enterprise LLM
Proprietary Fujitsu model
Specialization
Japanese language + Enterprise
Strong Japanese language capabilities
Key Feature
1-bit quantization
94% memory reduction, 89% accuracy retention
Release Date
2025
Ongoing development
Deployment
Government pilot
Japanese central government agency
Target Launch
FY2026
Generative AI service for policy/legislation
강점
- ・Exceptional Japanese language capabilities
- ・94% memory reduction via 1-bit quantization
- ・89% accuracy retention at extreme compression
- ・3x inference speedup
- ・Piloted in Japanese government operations
- ・AI-driven development platform covering full SDLC
약점
- ・Proprietary model, not publicly available
- ・Focused on Japanese market
- ・Limited public benchmark data
- ・Enterprise-only pricing
경쟁사 비교
| Model | Arena | SWE | GPQA | Price |
|---|---|---|---|---|
| ELYZA🐭 70B | - | - | - | Free (open) |
| Sarashina2 70B | - | - | - | Free (open) |
| GPT-4o Japanese | - | - | - | Paid API |
Takane is Fujitsu's enterprise LLM with exceptional Japanese language capabilities. Piloted in a Japanese central government agency for public comment operations, it features extreme compression (1-bit quantization with 94% memory reduction) and is being developed as a generative AI service for policy formulation and legislation.
출처
분석 생성일: 2026-05-24