MULTI-AGENT MODELV2.0 SINCE SEP 11OPENAI-COMPATIBLE API272K PRICE THRESHOLDNO EU/EEA
Fugu Ultra API Pricing
Sakana Fugu Ultra is the quality-first tier of Sakana Fugu, a multi-agent system delivered as one OpenAI-compatible model API. It lists at $5/M input, $30/M output and $0.50/M cached input, with higher rates once context exceeds 272K. On September 11, 2026 the fugu-ultra id moved to v2.0 on the same card; the dated id in this page's URL, fugu-ultra-20260615, is Sakana's name for v1.0 and still answers. Orchestration tokens bill on top of what you send and receive. Its cheaper sibling, Fugu Max, arrived the same day.
Input - per 1M tokens
$5.00/M
$10/M when context >272K standard
Output - per 1M tokens
$30.00/M
$45/M when context >272K standard
Cached input - 90% off
$0.50/M
$1/M when context >272K -90%
Effective - agentic blend
$3.61/M
92/8 split - 82% cache
§ 01 / TERMINAL
Run the numbers.
Live calculator pre-loaded with the standard Fugu Ultra token rates. For prompts whose context exceeds 272K, use the FAQ long-context rates or the scenario marked long-context.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
CODING AGENT
Repo-wide code review
$0.517/task
KAGGLE
Competition experiment loop
$0.836/run
LONG CONTEXT
Patent landscape investigation
$0.933/brief
SECURITY
Scoped security assessment
$0.345/assessment
Estimate · fugu-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 492
Words 73
Tokens (estimated) 128 tokens
Cost as input · uncached $0.00064 USD
Cost as output · uncached $0.00384 USD
Cost as cached input $0.00006 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Fugu Ultra Current | $5.00 cache $0.50 | $30.00 | $3.61 agentic 92/8 | 1M | Hard multi-agent orchestration |
| Fugu Max | $2.00 cache $0.25 | $6.00 | $1.00 −3.6× | 1M | Sakana's cost-first orchestration tier |
| GPT-6 Astra | $10.00 cache $1.00 | $50.00 | $6.82 +89% | 1.05M | Closed flagship Fugu Ultra v2 leaves out of its pool |
| Claude Fable 5.1 | $10.00 cache $0.25 | $50.00 | $6.26 +74% | 1M | Closed flagship Fugu Ultra v2 leaves out of its pool |
| Claude Opus 5.5 | $4.00 cache $0.20 | $20.00 | $2.58 −28% | 1M | Closed frontier agent work |
| GPT-6.1 Sol | $2.00 cache $0.10 | $10.00 | $1.29 −2.8× | 1.05M | Closed frontier coding at a lower rate |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 −7.5× · promo | 1M | Lower-cost multimodal context |
Compare adjacent tiers
Frequently asked.
Fugu Ultra has a fixed public token row, but Sakana Fugu itself is a multi-agent model API, so the page separates published prices from orchestration behavior and regional limits.
Q · 01 How much does Fugu Ultra cost? +
$5/M input, $30/M output and $0.50/M cached input on Sakana's pay-as-you-go card, which since September 11, 2026 names
fugu-ultra-v2.0. Above 272K of context the whole card steps up to $10 / $45 / $1.00. Checked on 2026-09-24.Q · 02 What happens above 272K context? +
The official pricing row lists higher rates when context exceeds
272K: $10/M input, $45/M output, and $1.00/M cached input. AI//COST keeps those as long-context rates instead of blending them into the headline price.Q · 03 Why can a request cost more than its input and output tokens suggest? +
Because orchestration is billed. Fugu Ultra's usage object carries
orchestration_input_tokens, orchestration_input_cached_tokens and orchestration_output_tokens, and Sakana's pricing page says these are real usage outside the input and output counts, charged at the same rates. The agents Fugu coordinates read and write tokens you never see, so budget from the usage object of real requests. Our calculator prices only the visible input and output.Q · 04 Is Fugu Ultra just a wrapper around other models? +
Sakana describes Fugu as a multi-agent system delivered as one model. It dynamically orchestrates a pool of expert models behind one OpenAI-compatible API, so AI//COST treats Fugu Ultra as a priced model product while noting the orchestration layer.
Q · 05 Is Sakana Fugu available in the EU or EEA? +
No. The official page says Sakana Fugu is not yet available in the EU/EEA while Sakana works toward GDPR and EU-specific regulatory compliance.
Q · 06 How is regular Fugu priced? +
Regular Fugu does not have a single fixed public input/output row. Sakana says one active agent is charged at the standard rate for that underlying model, and multiple active agents are charged as a single rate based on the top-tier model involved.
Q · 07 Which version does fugu-ultra serve now? +
v2.0, since September 11, 2026. Sakana's models page reads "fugu-ultra: The Fugu Ultra model, defaults to fugu-ultra-v2.0", and still lists
fugu-ultra-v1.1 and fugu-ultra-v1.0 — the latter "also known as fugu-ultra-20260615", the dated id in this page's URL. The pricing page now prints its fixed card for v2.0 only and gives no separate rate for the older versions. Sakana says v2 has a training cutoff of August 28, 2026 and that Claude Fable 5, Fable 5.1 and GPT-6 Astra are not in its pool.Q · 08 Does this page include subscription plans? +
The model calculator uses pay-as-you-go token pricing. Sakana also lists monthly plans, but those are account-plan allowances rather than direct per-token prices, so they are summarized on the Sakana provider hub instead of used as token math.
Q · 09 Fugu Ultra or Fugu Max? +
Fugu Max, released the same day as Ultra v2, costs
$2 input and $6 output per million at any context length. Sakana describes the two as one orchestration architecture tuned for different goals: Max for the best output at the lowest cost, Ultra for the highest capability. Ultra earns its premium only on work where Max's answers fail.