What Fugu-Cyber Is (a cyber-defense orchestration AI)
Fugu-Cyber at a glance
Let's start with what kind of model this is and how it differs from the original Fugu.
One API that coordinates specialized agents
Fugu-Cyber is a cyber-defense-focused orchestration AI model from Sakana AI. Orchestration means conducting multiple specialized AIs together, like a conductor leading an ensemble. You send a request to a single API endpoint, and internally a pool of specialized agents dynamically combines to handle complex, multi-step work. The defining trait of Fugu-Cyber is that it looks like a single model but is actually a system of specialized agents working in concert.
"Like our original Fugu orchestration model, Fugu-Cyber is a multi-agent system that behaves like a single model. You send a request to one endpoint, and the system dynamically orchestrates a pool of specialized agents to tackle complex, multi-step tasks without the risk of single-vendor dependency." — From the introduction
How it differs from the original Fugu
Sakana AI has long offered a general-purpose Fugu orchestration model. Fugu-Cyber is a variant purpose-built to fit the complexities of modern cyber defense. It keeps the vendor-independent design while aiming squarely at security work such as verifying vulnerabilities and processing threat intelligence.
"Available as a new API endpoint, Fugu-Cyber is purpose-built for the complexities of modern cyber defense." — From the introduction
Benchmark Performance and How to Read It
Fugu-Cyber benchmark scores (reported by Sakana AI)
Let's look at what the reported numbers mean and how far to trust them.
CyberGym 86.9%, CTI-REALM 72.1% (vendor-reported)
Sakana AI reported that Fugu-Cyber set top-tier scores on two benchmarks considered among the industry's hardest. CyberGym measures the ability to analyze complex codebases and verify real-world vulnerabilities, where Fugu-Cyber scored 86.9%. CTI-REALM measures turning raw threat intelligence into practical detection rules, where it scored 72.1%. The company calls this comparable to specialized models such as GPT-5.5-Cyber and Anthropic's Mythos-Preview.
"Fugu-Cyber achieves state-of-the-art performance on the industry's most challenging security benchmarks, reaching a success rate of 86.9% on CyberGym and 72.1% on CTI-REALM, comparable to leading cybersecurity-focused frontier models such as GPT-5.5-Cyber and Mythos-Preview." — From the introduction
A note: the numbers predate independent verification
What is worth keeping in mind is that these figures are all reported by Sakana AI itself and have not been independently verified by a third party. The details of the benchmark methodology are also not public at this point. The scores are a useful signal, but for now they are best treated as vendor-reported figures. If you are evaluating adoption, it makes sense to view them as something to confirm in your own environment.
Access and How to Use It (application and review)
Getting access to Fugu-Cyber
Fugu-Cyber is not available to everyone right away. Let's confirm the conditions.
The Token Plan, application, and manual review
Fugu-Cyber is offered through a new API endpoint at sakana.ai/fugu. Pricing sits within Sakana AI's Token Plan. Access first requires submitting an application form describing your intended use and verified contact information, and the company reviews and approves each one manually before granting access. Specific pricing is not stated publicly.
"The Fugu-Cyber API will be available for the Token Plan. Users will also need to apply for access by submitting an access request form detailing their intended use case and providing verified contact information. Our team will manually rigorously review and approve each application before granting users access to Fugu-Cyber." — From the Responsible Deployment section
A usage policy that bans offensive misuse
Behind the review process is the reality that an AI good at cyber defense can also be turned to attack. Fugu-Cyber was released under an updated Acceptable Usage Policy that prohibits offensive misuse. Because a tool powerful enough to help defenders also carries misuse risk, access is narrowed through review and policy.
"Fugu-Cyber is being released under an updated Acceptable Usage Policy that prohibits offensive misuse and aligns with the industry's safety standards." — From the Responsible Deployment section
Summary: How to Read Fugu-Cyber
Fugu-Cyber is a Japan-born orchestration AI model that pools specialized AIs for cyber defense. Its reported benchmarks are strong, but all are vendor-reported and independent verification is still ahead. The vendor-independent design and the review and policy meant to prevent offensive misuse are meaningful choices for bringing defensive AI into real operations. If you are weighing adoption, the practical path is to apply while planning to evaluate it in your own environment rather than taking the reported numbers at face value.
