Sakana Fugu: “One Model to Command Them All.” Overall, Sakana AI positions itself with “Building Frontier AI in Japan.”
Sakana AI develops frontier AI products and research projects. Fugu combines multiple external leading models behind an OpenAI-compatible API and orchestrates them dynamically for complex, multi-step tasks. Marlin is an autonomous research agent for market, technology, competitive, and regulatory analyses; Sakana Chat is a chat service based on the Namazu model family in the alpha phase.
Sakana AI
One Model to Command Them All.
Pro Monthly plan for regular coding, review, research, and analysis tasks; includes a higher usage allowance.
Max Monthly plan for longer, compute-intensive, and frequent multi-agent tasks. Other No permanently free production version
Credits must be purchased for API usage; an official free-tier version is not specified.
Token Plan / Pay-as-you-go Usage-based API access for production workloads and peak loads; billing based on token consumption.
Fugu Standard model focused on response speed, coding, code review, interactive work, and optional exclusion of individual model providers.
Fugu Ultra Model for demanding multi-step tasks, research, paper reproduction, patent and literature research; the full agent pool is predefined.
Enterprise Deployment Custom enterprise agreement for access, plans, and enterprise deployment; public details are unavailable.
Target audience
Sakana AI is aimed primarily at developers, AI research teams, corporate strategists, consultancies, financial and industrial companies, as well as public institutions in Japan. Fugu is intended for technically integrated multi-model workflows, while Marlin is designed for business research, market monitoring, strategy development, and management reports. For European organizations, the main relevance at present is research value, not productive Fugu deployment.
Outstanding features
Fugu abstracts multiple external top-tier models behind an OpenAI-compatible endpoint. The system dynamically decides on agent selection, roles, and model routing; individual providers can be excluded from the pool. Marlin takes a different approach: over several hours, the agent carries out hypothesis formation, web research, cross-checking, and strategic structuring, and delivers reports as well as presentation slides.
Main application areas
Useful applications include coding, code review, patent and literature research, market and competitive analysis, technology scouting, regulatory monitoring, strategic decision-making foundations, and scientific AI research. The AI Scientist and its successors are research projects for automated scientific work and should not be understood as general corporate SaaS.
Usage & notes
Fugu is integrated via API key through OpenAI-compatible Chat Completions and Responses endpoints. For German companies, this is currently not a standard product path because, according to the terms, Fugu is not offered in the EEA. Neither personal data nor trade secrets, health, financial, or other sensitive information should be processed in Fugu. Marlin is suitable only for publicly available information sources; the provider explicitly names purely internal, non-public datasets as an unsuitable scenario.
| Target audience | Assessment |
|---|---|
| Private individuals in Germany / EU | No – Sakana Fugu is currently not available for EU/EEA users. (Sakana AI) |
| Self-employed / freelancers outside the EU | Conditionally – interesting for coding, research, and complex analysis tasks; API knowledge is required. (Sakana AI) |
| SMEs outside the EU | Conditionally – suitable for AI-supported development, analysis, or chatbot applications, provided data protection and model provider risks are acceptable. (console.sakana.ai) |
| Large enterprises outside the EU | Yes, with enterprise review – Fugu offers usage-based API access and refers enterprise deployments to direct contact. (Sakana AI) |
| Developers / AI teams | Very well suited – core target group for OpenAI-compatible API integration, coding, code review, agentic workflows, and model orchestration. (Sakana AI) |
| Research / Data Science | Very well suited – Fugu Ultra is positioned for demanding research, paper reproduction, Kaggle tasks, patent and literature analysis. (Sakana AI) |
| Cybersecurity teams | Conditionally – according to the provider, usable for security analyses; results must be professionally reviewed and used only in authorized environments. (Sakana AI) |
| Privacy-sensitive organizations in Germany | No – no EU availability, external LLM routing architecture, and no public EU hosting/DPA information. (console.sakana.ai) |
| Model | API ID | Classification |
|---|---|---|
| Sakana Fugu | fugu | Standard model for coding, code review, chatbots, and interactive tasks; balances quality and latency. |
| Sakana Fugu Ultra | fugu-ultra | Performance variant for complex, multi-step tasks; uses a deeper expert agent pool and is slower and more expensive. The currently documented versioned Ultra variant is fugu-ultra-20260615. |
Hosting & Data
1) On-prem / local hosting
Meaning: The company operates the solution on its own hardware or within its own infrastructure. In the strictest sense, not only the application runs locally, but ideally the model as well.
2) Private cloud / data center
Meaning: The solution runs in a dedicated or more clearly separated cloud environment, often with a hosting provider or hyperscaler, but in a German data center or in a particularly controlled environment.
3) EU SaaS / managed
Meaning: The provider operates the solution itself as a service. The company uses the tool as a ready-made cloud service, ideally with EU data residency.
4) Hybrid
Meaning: One part of the processing remains internal / local / in a private cloud, while another part runs in an external cloud or EU SaaS.
5) AVV / DPA
Meaning: This is the data processing agreement or Data Processing Addendum. It governs that the provider processes personal data on behalf of the customer and is bound by the customer's instructions.
6) No training
Meaning: The provider does not use your prompts, uploads, attachments, chat histories, or outputs for training or improving the general model — ideally excluded by contract.
7) Open-source / transparency path
Meaning: There is a path toward greater technical transparency and sovereignty, for example through:
- open models
- documented components
- self-hostable parts
- traceable architecture
- export / switching options
| On-prem / local hosting | ❓ |
| Private cloud / data center | ❓ |
| EU SaaS / Managed | ❓ |
| Hybrid | ❓ |
| DPA / AVV | ❓ |
| No training on customer data | ⚠️ |
| Open source / transparency path | ⚠️ |
On-Prem / local hosting: indirect / not available
No on-prem, local, or self-hosted deployment for Fugu is specified on the website.
Private Cloud / data center: unclear
No dedicated private cloud, VPC, or segregated EU/EEA data center option is described on the website.
EU SaaS / Managed: unclear
Sakana operates Fugu as a cloud service via its own API, but EU/EEA data residency or an EU/EEA data center is not specified on the website.
Hybrid: indirect / not available
No hybrid deployment is described on the website in which one part is processed locally/private cloud and one part externally.
DPA / DPA: unclear
No DPA page, no published Data Processing Agreement process, and no corresponding contractual documents were found on the website.
No training: partial
The privacy policy states that personal data may be used to 'train, fine-tune, evaluate, and improve our models and systems,' but 'subject to available opt-out settings.' A possible opt-out is therefore provided, but not as a general default and, according to the text, possibly not retroactively.
Open source / transparency path: partial
The website describes Fugu as orchestration of multiple external models and documents API behavior as well as provider control via the 'custom model pool.' However, a genuine open-source, self-hostable, or fully transparent sovereignty path for the service itself is not specified on the website.
Data processing
The website describes Fugu as a managed API service at 'https://api.sakana.ai'. According to the privacy policy, personal data may include, among other things, prompts, uploads, outputs, technical data, and support data. The privacy policy also states that data may be disclosed to authorized vendors and service providers including LLMs as well as cloud/infrastructure providers. The Get Started section also states that 'fugu' routes across all supported providers by default and that providers can be excluded via the 'custom model pool.' Specific EU/EEA data residency, a subprocessors list, and data center locations are not specified on the website.
Conclusion
For an EU/EEA tool directory, the information available on the provider's website is currently too incomplete for a positive GDPR assessment. There is privacy documentation and an indicated opt-out path for training, but no clear information on EU/EEA hosting, DPA, subprocessors, or certifications. Since international transfers to Japan and the USA are also mentioned, GDPR-compliant use for EU/EEA users cannot be clearly substantiated based on the website alone.
Sources
| On-prem / local hosting | ❓ |
| Private cloud / data center | ❓ |
| EU SaaS / Managed | ❓ |
| Hybrid | ❓ |
| DPA / AVV | ❓ |
| No training on customer data | ⚠️ |
| Open source / transparency path | ⚠️ |
On-Prem / local hosting: indirect / not available
No on-prem, local, or self-hosted deployment for Fugu is specified on the website.
Private Cloud / data center: unclear
No dedicated private cloud, VPC, or segregated EU/EEA data center option is described on the website.
EU SaaS / Managed: unclear
Sakana operates Fugu as a cloud service via its own API, but EU/EEA data residency or an EU/EEA data center is not specified on the website.
Hybrid: indirect / not available
No hybrid deployment is described on the website in which one part is processed locally/private cloud and one part externally.
DPA / DPA: unclear
No DPA page, no published Data Processing Agreement process, and no corresponding contractual documents were found on the website.
No training: partial
The privacy policy states that personal data may be used to 'train, fine-tune, evaluate, and improve our models and systems,' but 'subject to available opt-out settings.' A possible opt-out is therefore provided, but not as a general default and, according to the text, possibly not retroactively.
Open source / transparency path: partial
The website describes Fugu as orchestration of multiple external models and documents API behavior as well as provider control via the 'custom model pool.' However, a genuine open-source, self-hostable, or fully transparent sovereignty path for the service itself is not specified on the website.
Data processing
The website describes Fugu as a managed API service at 'https://api.sakana.ai'. According to the privacy policy, personal data may include, among other things, prompts, uploads, outputs, technical data, and support data. The privacy policy also states that data may be disclosed to authorized vendors and service providers including LLMs as well as cloud/infrastructure providers. The Get Started section also states that 'fugu' routes across all supported providers by default and that providers can be excluded via the 'custom model pool.' Specific EU/EEA data residency, a subprocessors list, and data center locations are not specified on the website.
Conclusion
For an EU/EEA tool directory, the information available on the provider's website is currently too incomplete for a positive GDPR assessment. There is privacy documentation and an indicated opt-out path for training, but no clear information on EU/EEA hosting, DPA, subprocessors, or certifications. Since international transfers to Japan and the USA are also mentioned, GDPR-compliant use for EU/EEA users cannot be clearly substantiated based on the website alone.
Sources
Strengths & weaknesses at a glance
| Strengths | Weaknesses |
|---|---|
| • Multi-agent orchestration via an API | • Fugu is explicitly not available in the EU/EEA, UK, or Switzerland. |
| • OpenAI-compatible endpoints for existing integrations | • Fugu forwards inputs to external LLMs such as OpenAI, Anthropic, and Google. |
| • Strong for coding, code review, complex research, and multi-step analyses | • Fugu’s standard terms allow model training on content and possible human review; the Privacy Policy only refers to available opt-out settings. |
| • Marlin for in-depth market, competitive, and strategy research | • According to Fugu’s Terms, personal, sensitive, and confidential data must not be entered. |
| • Research strength through projects like The AI Scientist, Darwin Gödel Machine, and ShinkaEvolve | • No publicly discoverable DPA and no EU region secured for Fugu. |
| • Partially open-source research software available |
Reviews
0 reviews in total
There are no confirmed reviews for this tool yet.
Submit review
Your review will only become visible after email confirmation. This protects the portal against abuse.
Report review
Please select the reason why this review should be checked.
GDPR-compliant usage possible?
The website contains a privacy policy for 'Fugu', and it includes information on international data transfers, data subject rights, and a possible opt-out for training. However, for a reliable GDPR assessment across the entire EU/EEA, key information is missing from the website: a clearly specified EU/EEA server location or EU data residency, published DPA/AVV information, a subprocessor list, and specific details on certifications. In addition, the privacy policy explicitly mentions transfers to Japan and the USA. Therefore, GDPR-compliant use for EU/EEA users cannot be clearly demonstrated based on the website alone.
Positive
Positive aspects include the published privacy policy, the description of data subject rights, the reference to 'appropriate safeguards' for international transfers, and the statement that there may be an opt-out control for training/fine-tuning via account or product settings, if offered. In addition, according to the documentation, 'fugu' can restrict provider selection via the 'custom model pool', which at least suggests some degree of technical control over the disclosure to external model providers.
Negative
Negative is that the website does not guarantee an EU/EEA server location or EU data residency. Instead, the privacy policy states that personal data may be transferred to third parties outside the country of residence, including Japan and the USA. Also not specified on the website are a DPA/AVV page, a subprocessor list, on-prem/self-hosting options, and relevant certifications such as ISO 27001 or SOC 2.
Server location
No specific server/data center location for the EU/EEA is stated on the website. The privacy policy merely states that personal data may be transferred to third parties outside the country of residence, including Japan and the USA. EU data residency is not specified on the website.
