Published rates rechecked September 28, 2026
Estimate pay-as-you-go cost for the two Fugu models with a public rate table, without hiding orchestration usage or tool-call fees. The calculator excludes the base Fugu model because its route-dependent billing is not a single public number. It also excludes Fugu Cyber because official pages now route Cyber pricing to sales instead of publishing token rates.
Use the token counts from an API response when possible. API input totals already include their cached tokens, so enter the full total and the cached part inside it. Orchestration fields are extra usage outside those ordinary request totals. Fugu Max adds a per-call tool fee.
API usage reports cached tokens inside the matching input total, so the estimate bills uncached input — (input − cached) + (orchestration input − orchestration cached) — at the input rate, and both cached counts at the cached rate. A cached count larger than its input total is rejected instead of producing a doubled or negative charge. Leaving the orchestration fields at zero calculates only the visible request and response.
Ultra charges one rate up to 272K request context and a higher rate above it, judged per request rather than per month. An aggregate of many requests is not one long-context request, so pick the tier that matches the traffic you are counting. Fugu Max publishes a single rate at every context length, so its tier does not change the result. Split mixed Ultra traffic into two calculations, then add the results.
Fugu Max adds $0.007 for each web_search or web_fetch call on top of tokens. No per-call rate is published for Ultra, so the calculator rejects a non-zero tool-call count instead of guessing. Fugu Cyber remains pay-as-you-go after access, and its rates are now a sales conversation rather than a public table.
(((input − cached) + (orchestration input − orchestration cached)) / 1M x input rate) + ((cached + orchestration cached) / 1M x cached rate) + ((output + orchestration output) / 1M x output rate) + (tool calls x per-call rate)
Input totals are inclusive, so each cached count is a subset of its matching input total and is subtracted before the input rate applies; those tokens are billed once, at the cached rate. Orchestration fields stay extra usage outside the ordinary request totals. The tool-call term is zero unless the model publishes a per-call rate. Every entered value must be a whole number of tokens or calls between 0 and 9,007,199,254,740,991; a cached count above its input total, or any other invalid value, is rejected with a message instead of silently reusing an old result.
Subscriptions of $20, $100, and $200 are for the API platform only and do not describe an exact token allowance, so this tool does not compare subscription value. Fugu Max's published flat rate is not a quality claim about the base Fugu model.
Source: Sakana AI console pricing, rechecked September 28, 2026 for fugu-ultra-v2.0 and fugu-max-v1.0. Confirm the live page before purchasing because rates and availability can change.