Rules checked against first-party pages on September 28, 2026
Choose among Fugu, Fugu Ultra, Fugu Max, and Fugu Cyber
Answer three factual questions. The result follows the visible rule order below and can also say that no managed Fugu model is the safe starting point.
What this tool will return
A starting option, the exact reasons that fired, and a reminder to verify it on your own workload.
The complete decision order
- Region first. Official product pages say Fugu is currently unavailable in the EU and EEA, so the tool stops instead of naming a model.
- Auditability before performance. If every internal route must be visible, compare a self-managed graph before using a managed pool.
- Defensive security. Fugu Cyber is the documented specialist, but it requires access approval and is pay-as-you-go only.
- Provider exclusions. Standard Fugu lets you remove specific agents from its pool; Ultra uses a fixed full pool.
- Quality-first work. Ultra uses a fixed, deeper agent pool; current docs state no exact agent count.
- Cost-sensitive throughput. Fugu Max publishes one flat per-token rate at every context length, lower than Ultra's standard rate. The pricing page makes no quality claim for it, so benchmark it before treating it as an Ultra replacement.
- General default. Start with standard Fugu, then keep it only if a blind task evaluation justifies the cost and latency.
Sources: Sakana AI model catalog and Sakana AI product page.
What is deliberately not in the score
Subscription value is not scored because Sakana AI does not publish an exact token allowance for each tier. The cost-sensitive rule uses only the published per-token rate for Fugu Max, never a quality or allowance claim. Vendor benchmark ranks are not treated as your expected result. The chooser also does not infer legal or security authorization from a dropdown.