MULTI-AGENT MODELFLAT RATE AT ANY CONTEXT$0.007 PER WEB CALLOPENAI-COMPATIBLE APINO EU/EEA
Fugu Max API Pricing
Fugu Max is the cost-first half of Sakana's September release: one OpenAI-compatible model that routes each task across Sakana's largest pool of open-weight and specialised models, NVIDIA's Nemotron family among them. It lists at $2/M input, $6/M output and $0.25/M cached input, and unlike Fugu Ultra the card does not step up above 272K: Sakana prices it regardless of context length. What the card leaves out is web access. Search and page fetches bill at $0.007 per call, and Sakana warns that one query can take several.
Input - per 1M tokens
$2.00/M
Fugu Ultra $5/M any context
Output - per 1M tokens
$6.00/M
Fugu Ultra $30/M any context
Cached input - per 1M tokens
$0.25/M
No step above 272K 88% off
Effective - agentic blend
$1.00/M
92/8 split - 82% cache
§ 01 / TERMINAL
Run the numbers.
Calculator pre-loaded with Fugu Max's pay-as-you-go rates. It counts tokens only: add $0.007 for every web search or page fetch the model makes, which a single research question can multiply.
$ /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
$0.140/task
LONG CONTEXT
Contract bundle review
$0.286/bundle
RESEARCH
Market research brief
$0.070/brief
RAG SUPPORT
Support agent ticket triage
$0.011/ticket
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 445
Words 72
Tokens (estimated) 116 tokens
Cost as input · uncached $0.00023 USD
Cost as output · uncached $0.0007 USD
Cost as cached input $0.00003 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Fugu Max Current | $2.00 cache $0.25 | $6.00 | $1.00 agentic 92/8 | 1M | Multi-agent orchestration priced for volume |
| Fugu Ultra | $5.00 cache $0.50 | $30.00 | $3.61 +3.6× | 1M | Sakana's quality-first orchestration tier |
| Sakana Namazu | $0.95 cache $0.15 | $4.00 | $0.59 −41% | 256K | Japanese-language work on a single model |
| Claude Sonnet 5.5 | $2.00 cache $0.10 | $10.00 | $1.29 +29% | 1M | Closed frontier agent and coding work |
| GPT-5.6 Terra | $2.00 cache $0.20 | $12.00 | $1.52 +53% | 1.05M | Mid-tier closed model for everyday agents |
| GPT-6.1 Sol | $2.00 cache $0.10 | $10.00 | $1.29 +29% | 1.05M | Closed frontier coding at a lower input rate |
| DeepSeek V4 Pro | $1.32 cache $0.044 | $3.96 | $0.569 −43% | 1M | Frontier reasoning with a 1M context |
Frequently asked.
What Fugu Max costs, what the token rate does not cover, and how it differs from Fugu Ultra.
Q · 01 How much does Fugu Max cost? +
$2/M input, $6/M output and $0.25/M cached input, pay-as-you-go. Sakana's pricing page says the card applies regardless of context length, so there is no long-context step. Under our 92/8 agentic blend with 82% cache hits the planning figure is $1.00/M.
Q · 02 What does the token rate not cover? +
Web access.
web_search and web_fetch bill at $0.007 per call, and Sakana explains that because the open models in the pool have no native search, Fugu Max runs its own internal web tools, where a single query may need several calls. A research question that triggers ten searches and fetches adds $0.07 before any tokens. Sakana's pricing page also documents separately billed orchestration tokens for Fugu Ultra and Fugu Cyber; it says nothing either way about Fugu Max, so check the usage object of your own requests.Q · 03 Fugu Max or Fugu Ultra? +
Fugu Ultra costs 2.5x more per input token and 5x more per output token, and its rates double on input and rise by half on output once context passes 272K; Fugu Max stays flat. Sakana describes them as the same orchestration architecture tuned for two goals: Max for the best output at the lowest cost, Ultra for the highest capability. Ultra runs a deeper fixed pool of expert agents; Max runs the largest pool of open-weight and specialised models. Start on Max and move a workload to Ultra only where Max's answers fail.
Q · 04 Is it really 40-60% cheaper than Sonnet 5, GPT-5.6 Terra and Kimi K3? +
On output, yes: Sakana's claim compares output prices, and $6 against Claude Sonnet 5's $10, GPT-5.6 Terra's $12 and Kimi K3's $15 is 40%, 50% and 60% lower. Input is a different story: Sonnet 5 and GPT-5.6 Terra also charge $2 per million input tokens, and in agent work input is usually 90% or more of the tokens. The benchmark wins in the launch post are Sakana's own measurements, including an internal benchmark.
Q · 05 Which models does Fugu Max route to? +
Sakana does not publish the list. It says Fugu Max orchestrates its largest pool yet of open-weight and specialised models, including NVIDIA's Nemotron family through a partnership announced in August 2026, and that it routes each task to the leanest model able to solve it. For enterprise requests that need specific models or providers, Sakana asks you to contact it.
Q · 06 What is the context window? +
1M tokens, with text and image input. The pricing page does not state a window; the figures come from Sakana's own Codex setup guide, whose model catalogue lists
fugu-max with context_window: 1000000 and input_modalities: ["text", "image"]. Unlike Fugu Ultra, the price does not change anywhere inside that window.Q · 07 Where is Fugu Max available? +
Through Sakana's OpenAI-compatible API, including the Responses, Chat Completions and Models endpoints and an Anthropic-compatible Messages endpoint, and inside all three monthly plans (Standard $20, Pro $100, Max $200). Sakana Fugu is not available in the EU or EEA while Sakana works toward GDPR compliance. The model id is
fugu-max, which currently points to fugu-max-v1.0.