Fugu Max and Ultra v2.0 landed with flat published rates
The source-check log lists every first-party fact that changed or stayed the same this recheck — the new fourth model, the Ultra ID move, and the API subscription versus free Chat split.
Independent guide, source checked September 2026
Sakana Fugu: what it is, what it costs, and whether it's free
Sakana Fugu is Sakana AI's multi-agent model family with an OpenAI-compatible API. Fugu is the fast default, Fugu Ultra adds deeper reasoning, Fugu Max targets cost-performance, and Fugu Cyber covers defensive security. The API has no permanent free tier — paid API subscriptions start at $20/month with pay-as-you-go billing — while Sakana Chat, the separate consumer product, stays free with its own usage limits. See the pricing guide or calculate your cost.
Sakana Fugu Quick Answers
- Free & cost: The API has no free tier — paid API subscriptions start at $20/month, and pay-as-you-go is available. Sakana Chat is a separate free product with its own usage limits.
- Model Lineup: Fugu (fast default), Fugu Ultra (quality-first reasoning & code), Fugu Max (largest pool, cost-performance), and Fugu Cyber (defensive security).
- API Integration: Standard OpenAI SDKs connect directly using custom
baseURLand your Sakana API key.
From the library
Start with the decision that is actually on your desk
What this site can and cannot verify
Product names, endpoint support, model IDs, rates, and published scores are checked against Sakana AI first-party pages. We do not claim private access or unpublished hands-on results. Benchmark numbers below are Sakana AI reported results unless a separate independent source is named.
Fast orientation
What Sakana Fugu is, in one minute
Think of Fugu as a project lead behind one model name. A request can be answered directly or routed across a small team, then synthesized into one response. The orchestration is learned rather than defined by a workflow you maintain. For the operational view — what the agent does, how orchestration is billed, and which model to pick — read the dedicated Fugu AI guide.
Important distinction:
Fugu is not the same product category as OpenRouter or a self-managed LangGraph pipeline. OpenRouter gives broad model access, while a self-managed graph gives you explicit routing. Fugu's main trade is less orchestration work in exchange for less control over the learned route.
Model selection
Fugu vs Fugu Ultra vs Fugu Max
The right starting point depends on task shape, not which name sounds strongest. Fugu Cyber is a separate, access-gated security model noted just below.
Fugu
Best first test for interactive coding, chat, and routine work where response time matters.
- Balances quality and latency
- Routes to an underlying model based on the task
- Supports provider or model opt-outs
- Pay-as-you-go cost is variable, not one fixed public rate
Fugu Ultra
Use when answer quality matters more than latency, then verify the gain on your own hard tasks.
- Coordinates a deeper, fixed agent pool
- No provider or model opt-outs
- Fixed $5 / $30 / $0.50 rates per 1M tokens
- Orchestration tokens are billable
Fugu Max
Coordinates the largest pool and trades some peak quality for a much lower published rate.
- Flat $2 / $6 / $0.25 rates per 1M tokens, any context length
- Also a fixed pool with no opt-outs
- Built-in web_search and web_fetch calls bill at $0.007 each
- A sensible default when budget, not peak quality, is the constraint
Fugu Cyber is access-gated.
Fugu Cyber targets authorized defensive security reasoning, vulnerability research, and threat investigation. It is pay-as-you-go only, needs a reviewed use-case access request, and Sakana AI no longer publishes Cyber token rates — the official pricing page now says to contact sales. Authorization and scope remain the user's responsibility.
Practical rule:
Start with Fugu for normal work. Try Fugu Max when cost-per-result matters and Fugu Ultra when peak quality does, then verify the trade on your own hard tasks. Consider Cyber only for authorized defensive security work and only after confirming access.
Use the transparent model chooserCurrent public rates
Pricing that includes the hidden-looking part: orchestration usage
Rates below are per one million tokens and were checked against the official console pricing page on September 28, 2026. Fugu Cyber no longer has a public rate table.
| Model | Input | Output | Cached input | Context over 272K |
|---|---|---|---|---|
| Fugu | Variable: underlying model rate or top-tier participating model rate | Depends on the routed model | ||
| Fugu Ultra (v2.0) | $5.00 | $30.00 | $0.50 | $10 / $45 / $1.00 |
| Fugu Max (v1.0) | $2.00 | $6.00 | $0.25 | Same rates regardless of context |
| Fugu Cyber | Not published on the official pricing page as of September 28, 2026. Contact Sakana AI sales after access is approved. | |||
What the simple token headline misses
Ultra and Cyber report orchestration input, cached input, and output fields in addition to user-visible input and final output. Sakana AI says these fields represent real usage and are charged at the same input, cached, or output rate. The cached counts are subsets of their own input totals — cached_tokens is already inside input_tokens — so they are subtracted before the full input rate applies, not added on top. A calculator that ignores the orchestration fields can materially understate a long multi-agent run, while one that double-counts cache overstates it.
Subscriptions are for the API, not Sakana Chat
Standard is $20 per month, Pro is $100, and Max is $200 — these are Sakana AI API Platform subscriptions only. Every tier includes Fugu, Fugu Ultra, and Fugu Max. Sakana Chat is a separate product with a free usage allowance and no subscription plan, so subscribing does not change its limits. The public page describes Pro as 10 times Standard usage and Max as 20 times, but it does not publish an exact token allowance, so treat the plan prices as budget references, not guaranteed token equivalents.
Source: Sakana AI console pricing, rechecked September 28, 2026. Fugu Max flat rates apply regardless of context length, and its built-in web_search and web_fetch calls bill at $0.007 each.
Integration path
The API is familiar, but not every OpenAI parameter behaves the same
Sakana AI supports the Responses, Chat Completions, and Models APIs. Its documentation recommends the Responses API for better tool, multimodal, reasoning, and function-call behavior.
Python, Responses API
import os
from openai import OpenAI
api_key = os.environ["FUGU_API_KEY"]
base_url = os.environ["FUGU_BASE_URL"].rstrip("/")
if not base_url.endswith("/v1"):
base_url += "/v1"
client = OpenAI(api_key=api_key, base_url=base_url)
response = client.responses.create(
model="fugu",
input="Review this function and list correctness risks first.",
)
print(response.output_text)Model IDs
Use fugu, fugu-ultra, fugu-max, or fugu-cyber. Versioned IDs such as fugu-ultra-v2.0 and fugu-max-v1.0 are also documented.
Ignored fields
Several sampling controls are accepted but ignored. Do not assume temperature, top_p, or penalty fields change behavior.
Bounded retries
Set a finite timeout and a small retry count. A multi-agent request can be long, but an unbounded retry loop can duplicate expensive work.
Usage capture
Persist the full usage object for cost review. Final input and output fields alone do not represent all billable orchestration work.
Evidence, not decoration
How to read the Sakana Fugu benchmark table
The published results are useful evidence about the vendor's test setup. They are not proof that Ultra will beat every frontier model on your repository, prompts, tools, latency budget, or failure modes.
What the published results support
- Sakana AI's summaries position Fugu and Ultra as competitive across several coding and reasoning suites.
- Ultra is presented as the peak-quality option and Fugu Max as the cost-performance option, and the two overlap on several benchmarks.
- Different tasks benefit from different orchestration depth.
What it does not establish
- Independent reproduction across all scores
- Lower latency or lower total cost
- Reliability on your private code and toolchain
- Full visibility into which agent caused a good or bad result
Use benchmarks to choose tests, not to skip tests.
Use the benchmark evaluation playbook
A fair evaluation uses the same prompts, tools, timeouts, scoring rubric, and repeat count for Fugu, Ultra, and your existing baseline. Record failures and costs, not only the best-looking response.
Deep guides
Twelve source-checked guides, one decision each
Every guide solves one concrete decision and links time-sensitive product facts back to a first-party source.
API quickstart and failure checklist
Choose a model ID, validate configuration, stream a response, understand ignored parameters, and keep retries bounded.
Pricing and orchestration token accounting
Reconcile the usage object, calculate Ultra and Max charges, and avoid making up a fixed Fugu rate.
Benchmark reading and evaluation playbook
Turn vendor scores into a blind, repeatable test on the work that actually matters to your team.
Failure checklist
Turn 401, missing models, EU unavailability, and unexpected Ultra bills into a next check.
Fugu Cyber access
Pay-as-you-go only, official request, unpublished rates, and the authorized-use boundary.
Fugu versus a graph you can inspect
When a learned pool is the right trade, and when routing visibility still requires self-managed orchestration.
Orchestration token accounting
Why Ultra and Cyber usage objects do not match OpenAI's token_details, and which fields to persist for a bill.
Subscription versus pay-as-you-go
What the $20 / $100 / $200 tiers disclose, and why a break-even formula is still dishonest.
Responses, Chat, and Messages surfaces
Three supported shapes, ignored sampling fields, and why previous_response_id is not conversation state.
Official API versus listed partners
What the product page lists for OpenRouter, Vercel, and opencode, and what this site still cannot verify.
EU and EEA availability
Official unavailability is a stop, not a network problem; this page records the rule.
Pin model IDs
When to use the moving fugu-ultra alias and when to pin v2.0, v1.1, or v1.0.
Decision checklist
A small evaluation that produces a defensible answer
- 1
Choose ten representative tasks
Include routine, difficult, ambiguous, and failure-prone work. Do not build a set made only of prompts that flatter orchestration.
- 2
Freeze the run contract
Use the same prompt, available tools, timeout, output constraint, and retry policy for every candidate.
- 3
Score output blind
Hide the model name while a reviewer scores correctness, evidence, completeness, and unnecessary claims.
- 4
Capture operational cost
Record wall time, failures, retries, and the entire usage object, including orchestration token details.
- 5
Adopt the narrow winner
Use Fugu where it wins. Try Fugu Max where cost-per-result matters and Ultra where the quality delta is worth the cost. Keep the existing tool where none earns a switch.
Verification log
What was rechecked on September 28, 2026
- Models: official pages now list four — Fugu, Fugu Ultra, Fugu Max, and Fugu Cyber.
fugu-ultradefaults tofugu-ultra-v2.0, and the newfugu-maxdefaults tofugu-max-v1.0. - New rates: Fugu Max publishes flat $2 / $6 / $0.25 per 1M tokens at any context length. Fugu Ultra holds at $5 / $30 / $0.50, rising to $10 / $45 / $1.00 above 272K.
- Cyber pricing: still no public rate table. The live pricing page says to contact sales.
- Subscriptions vs Chat: the $20 / $100 / $200 tiers are API Platform only and include Fugu, Ultra, and Max — not Cyber. Sakana Chat is separate, free, and unaffected by them.
- Web tools: Fugu Max routes web_search and web_fetch through internal tools billed at $0.007 per call.
- API surfaces: Responses remains recommended; Chat Completions and Anthropic Messages are also documented.
- Usage fields: orchestration input, cached input, and output remain separately reported and billable on Ultra and Cyber. Each cached count is a subset of its own input total, so it is subtracted before the full input rate is applied — this site corrected its earlier formula, which had added cached tokens on top of an input total that already contained them.
- Partners: the product page lists OpenRouter, Vercel, opencode, Creao, and Merge. Those listings are not a copy of official billing rules.
Common questions
Sakana Fugu FAQ
Is Sakana Fugu free?
The API is not free: there is no permanent free API tier, and paid API subscriptions start at $20 per month with pay-as-you-go billing alongside them. Sakana Chat, Sakana AI's separate consumer chat product, is free with its own usage allowance, and an API subscription does not change those limits.
Which model should I start with?
Start with Fugu for routine or interactive work. Try Fugu Max when cost per result matters most and Fugu Ultra when peak quality does, then compare them on a small set of difficult tasks. Use Cyber only for authorized defensive security work after confirming access.
Can I calculate Fugu pay-as-you-go cost in advance?
Not as one reliable Fugu fixed rate, because Fugu bills at the underlying model's rate or the top-tier model involved. Fugu Ultra and Fugu Max publish fixed per-token rates, so a fixed-rate calculator covers those; Cyber pricing is only available through Sakana AI sales.
Are the benchmark results independently verified?
The table is published by Sakana AI and some baseline values are provider reported. Treat it as vendor evidence. Our evaluation guide shows how to test the same decision on your own workload without pretending we ran unpublished experiments.
Is SakanaFugu.com operated by Sakana AI?
No. SakanaFugu.com is an independent editorial guide and has no affiliation, endorsement, or sponsorship relationship with Sakana AI.
Primary sources
- Sakana AI product page: models, positioning, pricing overview, FAQ, and regional availability
- Sakana AI model documentation: model IDs, endpoints, supported fields, benchmark table, and usage fields
- Sakana AI pricing documentation: fixed rates, Fugu billing behavior, and subscriptions
- Sakana Fugu Technical Report: training approach and reported evaluation methodology
Corrections are welcome at [email protected]. Please include a first-party source when possible.