Estimate what a Sakana Fugu Ultra or Sakana Namazu workload will actually cost. Unlike a plain price-per-token table, this calculator accounts for the two things that most often surprise new Fugu users: orchestration tokens (Fugu Ultra is a multi-agent system, and the tokens its internal agents consume are real, billed usage at standard rates) and the long-context price tier that kicks in above 272K tokens of context.
Everything runs in your browser — nothing you type here is sent to any server.
Orchestration overhead applies to Fugu models only — Sakana Namazu is a single reasoning model. Note: Namazu bills thinking tokens at the output rate, and its built-in tools (web search, code execution) are billed separately.
Estimated cost
| Input tokens | – |
| Cached input tokens | – |
| Output tokens | – |
| Orchestration tokens (est.) | – |
| Per request | – |
| Per month | – |
Rates as of August 2026, from Sakana AI's official pricing page — Fugu Ultra: $5 input / $30 output / $0.50 cached input per 1M tokens ($10 / $45 / $1.00 above 272K context); Sakana Namazu: $0.95 / $4.00 / $0.15. This is an unofficial estimate, not a quote — always verify current rates on console.sakana.ai/pricing before making decisions. All calculations run in your browser; nothing is transmitted.
How the estimate works
- Input tokens are billed at the model’s input rate; the share you mark as cached is billed at the much lower cached-input rate instead.
- Orchestration overhead models the extra tokens Fugu Ultra’s internal agents consume. Sakana’s pricing page states these are counted in the final price of the request, at the same rates as standard input and output tokens. Real overhead varies a lot by task — the slider is deliberately adjustable, with a mid-range default.
- Long-context tier: above 272K tokens of context, Fugu Ultra’s rates roughly double ($10 input / $45 output / $1.00 cached per 1M tokens).
- Sakana Namazu is a single (non-orchestrating) Japanese-specialized model, so the overhead slider doesn’t apply. Its thinking tokens are billed at the output rate, and its built-in tools cost extra (web search $7.00 per 1,000 calls, code execution $0.12 per hour, as of August 2026).
Want to try the models with your own key instead? Head over to the Fugu Playground.