Independent guide. Last checked June 22, 2026.
Sakana Fugu Guide: pricing, API, benchmarks, and alternatives
A decision tool for developers evaluating Sakana AI's multi-agent orchestration API.
Sakana Fugu in 30 seconds
What Is Sakana Fugu?
Sakana Fugu is not a sushi page and not a simple model marketplace. It is Sakana AI's productized orchestration layer: users call one API, while Fugu decides how to route, delegate, verify, and synthesize work across expert agents.
A model-shaped coordinator
Fugu behaves like a single model from the developer side, but its product promise is learned coordination across a model pool.
Best for messy work
It is most interesting for code review, research loops, security assessment, literature search, and tasks that need checking.
Not a magic default
Simple chatbot turns and cheap Q&A may be better served by a smaller direct model.
Fugu vs Fugu Ultra
Pick the workload that looks closest to yours. The recommendation is intentionally conservative.
Sakana Fugu Pricing
Official pricing includes monthly plans and pay-as-you-go. Treat benchmark and cost claims as live inputs to recheck before production.
Standard
$20/mo
Light personal usage. Includes both Fugu and Fugu Ultra.
Pro
$100/mo
10x Standard usage for regular coding, review, and research sessions.
Max
$200/mo
20x Standard usage for long-running and heavier tasks.
Estimate uses official per-1M-token Fugu Ultra rates.
Sakana Fugu Benchmarks
These are vendor-reported numbers from the official Sakana page. Use them as a starting point, not as a procurement proof.
| Benchmark | Fugu | Fugu Ultra | What it means |
|---|---|---|---|
| SWE Bench Pro | 59.0 | 73.7 | Coding and software engineering tasks |
| LiveCodeBench | 92.9 | 93.2 | Competitive coding style evaluation |
| GPQA-D | 95.5 | 95.5 | Graduate-level science questions |
| Humanity's Last Exam | 47.2 | 50.0 | Broad hard reasoning benchmark |
How to Use the Sakana Fugu API
The official site says Sakana Fugu is available through an OpenAI-compatible API. Keep the endpoint and model name configurable so you can swap between Fugu and Fugu Ultra without a code rewrite.
- Get access from the Sakana AI console.
- Set your base URL from the console, not from copied blog posts.
- Start with Fugu for latency, then escalate difficult tasks to Fugu Ultra.
- Log tokens, latency, failures, and answer quality before scaling.
Python quickstart pattern
from openai import OpenAI
import os
client = OpenAI(
api_key=os.environ["SAKANA_API_KEY"],
base_url=os.environ["SAKANA_BASE_URL"],
)
response = client.chat.completions.create(
model="fugu-ultra-20260615",
messages=[
{"role": "user", "content": "Review this pull request and list risks."}
],
)
print(response.choices[0].message.content)Best Use Cases for Sakana Fugu
Use Fugu where orchestration can reduce your decision load or add verification.
Code review
Large diffs, multi-file reasoning, regression risk, and test suggestions.
Research loops
Paper reading, experiment planning, reproduction attempts, and failure analysis.
Security assessment
Scoped reviews where evidence, retest steps, and restraint matter.
Patent and literature search
Broad retrieval, synthesis, contradiction checks, and citation triage.
Sakana Fugu vs OpenRouter
OpenRouter, LangGraph, CrewAI, and AutoGen solve adjacent problems. The key question is whether you want to buy orchestration or build it.
Sakana Fugu
Model-shaped orchestration API. Good when you want one endpoint and learned routing.
OpenRouter
Unified access to many models. Good when you want provider choice and explicit model selection.
LangGraph, CrewAI, AutoGen
Frameworks for building your own workflows. Good when control and observability matter more than speed.
Sakana Fugu Limitations
Do not integrate only because launch benchmarks look strong. Production fit depends on governance, latency, observability, and region support.
- Underlying routing details are not exposed like a self-built agent graph.
- Fugu Ultra can trade response time for deeper coordination.
- EU/EEA availability is blocked while compliance work continues.
- Simple tasks may be cheaper through direct model calls.
- Vendor-reported benchmarks need your own workload validation.
Sakana Fugu FAQ
Is Sakana Fugu related to pufferfish?
The name references fugu, but this guide targets Sakana AI's AI orchestration product, not food, recipes, or restaurants.
Should I use Fugu or Fugu Ultra?
Use Fugu when latency and cost matter. Use Fugu Ultra when difficult multi-step quality matters more than speed.
Can I see which model answered?
Do not assume full routing transparency. If exact provider audit logs are a hard requirement, compare self-built orchestration first.
Is SakanaFugu.com official?
No. This is an independent guide and is not affiliated with Sakana AI.