A generative AI platform launched by SAKURA internet in September 2025. Its defining feature is that model execution, network communication, and data storage are all completed within data centers in Japan, which makes it possible to add generative AI to workloads whose data cannot leave the country. It is offered in two parts: the inference API “Sakura no AI Engine” for developers, and “AI Solutions” for enterprises, which bundles implementation support. Because the API is OpenAI-compatible, existing SDKs and code can be reused almost as-is.
Key Features
- Data processing completed domestically: Model execution, communication, and storage all take place on cloud infrastructure in Japan, under Japanese law. Input data is not shared with the model providers and is not used for training
- OpenAI-compatible API: The interface is compatible with OpenAI and Anthropic, so migrating an existing generative AI app can be as simple as swapping the endpoint
- A model lineup organized by task: Alongside gpt-oss-120b for chat and the domestically developed llm-jp-3.1-8x13b-instruct4, a single API key also covers whisper-large-v3-turbo for speech recognition, multilingual-e5-large for embeddings, and VOICEVOX for speech synthesis
- Document management for RAG: A RAG foundation is provided for registering documents in chunks and using them for retrieval, so there is no need to build that layer yourself
- A Playground you can try in the browser: Responses from several models can be compared in the browser before you sign a contract or start implementing
- Security certifications and implementation support: The service holds the Privacy Mark, ISO27001, and PCI DSS. On the AI Solutions side, packages such as InfiniCloud AI, neoAI Chat, and meeting transcription tools come with implementation support
Pricing
There are two options ─ a free tier for the base models and a pay-as-you-go plan. The free allowance resets every month.
| Target | Free allowance (per month) | Pay-as-you-go rate |
|---|---|---|
| Chat (gpt-oss-120b / llm-jp-3.1-8x13b-instruct4) | 3,000 requests | ¥0.15 per 10K input tokens, ¥0.75 per 10K output tokens |
| Speech recognition (whisper-large-v3-turbo) | 50 requests | ¥0.5 per 60 seconds |
| Embeddings (multilingual-e5-large) | 10,000 requests | ¥2 per 10K tokens |
| Speech synthesis (VOICEVOX) | 50 requests | ¥3 per 10,000 morae |
| Document management (RAG) | None | ¥3 per 100 chunks |
Models offered as previews (Kimi-K2.6, Qwen3.6-35B-A3B, gemma-4-31B-it, and others) are priced differently from the table above. Pricing on the AI Solutions side varies by configuration, so you need to inquire through the official site.
Pricing is current as of August 2026. Please check the official site for the latest pricing.
Pros & Cons
✅ Pros
- Data never leaves the country and is not shared with model providers, which makes it easier to consider for work involving personal information or confidential documents
- With an OpenAI-compatible API, existing implementations and libraries can be carried over almost unchanged
- The free allowance resets monthly, so small validations and internal tools can be started at no cost
- Chat, speech recognition, embeddings, speech synthesis, and RAG are consolidated behind one contract, reducing the effort of stitching services together
- Holding the Privacy Mark, ISO27001, and PCI DSS makes it easier to satisfy internal procurement requirements
⚠️ Cons
- The lineup centers on open-weight models and domestic LLMs; the latest frontier models from the large overseas vendors are not available
- Some models are offered as previews, so specifications and prices may change
- Image generation is not part of the published model lineup, so a separate service is needed for creating images
- Document management (RAG) has no free allowance and is billed from the moment documents are registered
- Documentation and support are primarily in Japanese, which can be awkward for teams with many overseas members
Comparison with Similar Services
| Item | Sakura no AI | Amazon Bedrock | Azure OpenAI Service | OpenAI API |
|---|---|---|---|---|
| Provider | SAKURA internet (Japan) | Amazon (US) | Microsoft (US) | OpenAI (US) |
| Data location | Completed in domestic data centers | Region selection (Tokyo available) | Region selection (Japan available) | Regions designated by the provider |
| Main models | Open-weight plus domestic LLMs | A choice of models from several vendors | Mainly GPT-family | GPT-family |
| API format | OpenAI / Anthropic compatible | Proprietary SDK (with compatibility layers) | OpenAI-compatible | Standard |
| Free allowance | Yes (resets monthly) | Generally none | Generally none | Generally none |
Who Is It For
- Developers who need to embed generative AI in business systems that cannot send data abroad because they handle personal information or confidential documents
- Organizations in government, healthcare, or finance, where procurement requirements ask about data location and certifications
- Teams that already built an app on the OpenAI API and want to move it onto infrastructure that stays within Japan
- IT staff who want to set up a chatbot that references internal documents, RAG foundation included
- Small businesses and individual developers who want to try building generative AI features in-house within a free allowance
Summary
Sakura no AI is not aiming for the front line of the performance race; it answers the requirement of “using generative AI while keeping data inside the country” head-on. The OpenAI-compatible API and the monthly-resetting free allowance let you start small while reusing existing implementations. It is not suited to work that demands the performance of the newest frontier models, but it is a strong option where data location drives the decision.