sakana · Frontier
Fugu Max
Sakana AI's cost-performance Fugu model uses learned multi-agent orchestration to route tasks across expert models for reasoning, coding, and tool use.
- Reasoning
- Tools
- Vision
- Structured output
- PDF input
- Input
- Text, Images
- Output
- Text
- Reasoning effort
- high · xhigh · max
- Discovery tier
- Frontier
Pricing
USD per 1M tokens- Input
- $2.00
- Output
- $6.00
- Cache read
- $0.25
Reference prices as published by the model provider. API example
Specifications
- Context window
- 1M
- Max output
- 128K
- Average output (7 days)
- 2.2K tokens
- Longer than 92% of 473 models
Side by side
Fugu Max compared
Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.
| Fugu MaxsakanaThis model | Kimi K3.1 ProMoonshotRemove | |
|---|---|---|
| Provider | sakana | Moonshot |
| Discovery tier | Frontier | Frontier |
| Added | Sep 11, 2026 | Oct 24, 2026 |
| Limits | ||
| Context window | 1M | 1.05MBest |
| Max output | 128K | — |
| Average output | 2.2K tokens | 921 tokens |
| Pricing, USD per 1M tokens | ||
| Input | $2.00 | $2.00 |
| Output | $6.00Best | $10.00 |
| Cache read | $0.25 | $0.20Best |
| Cache write | — | $2.00 |
| Performance | ||
| Throughput | — | 42.1 tok/s |
| Time to first token | — | 6.7 s |
| Request success (24h) | 100.0%Best | 83.0% |
| Benchmark indexes | ||
| Intelligence Index | — | 55.1 |
| Capabilities | ||
| Reasoning | Yes | Yes |
| Tools | Yes | Yes |
| Vision | Yes | Yes |
| Structured output | Yes | Yes |
| PDF input | Yes | No |
| Video input | No | Yes |
| Reasoning effort | high, xhigh, max | low, high, max |
Public evaluations
Benchmarks
Independent evaluation results, with how each score ranks among the models in the Noumen catalog.
- Fugu Max
- Kimi K3.1 Pro
Intelligence Index
- Fugu Max—
- Kimi K3.1 Pro55.1
Agentic work
AutomationBench-AA
Workflow automation, penalized for guardrail violations
- Fugu Max: —
- Kimi K3.1 Pro: 73.3%
AutomationBench-AA tasks completed
Workflows finished end to end without a guardrail violation
- Fugu Max: —
- Kimi K3.1 Pro: 46.0%
AA-Briefcasev1.1
Agentic knowledge work, rated head to head
- Fugu Max: —
- Kimi K3.1 Pro: 1,678 Elo
GDPval-AAv2.1
Economically valuable occupational tasks, rated head to head
- Fugu Max: —
- Kimi K3.1 Pro: 1,846 Elo
Document reasoning
GDP.pdf
Professional reasoning over PDF documents, all-pass rate
- Fugu Max: —
- Kimi K3.1 Pro: 31.4%
AA-LCRv1.1
Reasoning across long documents
- Fugu Max: —
- Kimi K3.1 Pro: 86.3%
Reasoning and knowledge
Humanity’s Last Exam
Expert-level questions across academic fields
- Fugu Max: —
- Kimi K3.1 Pro: 52.5%
Coding
Terminal-Benchv4.0
Practical software engineering in a terminal
- Fugu Max: —
- Kimi K3.1 Pro: 67.2%
SciCode
Scientific computing in Python
- Fugu Max: —
- Kimi K3.1 Pro: 60.8%
Legacy evaluations
AA-LCR (unversioned)
Long-context reasoning, earlier grading
- Fugu Max: —
- Kimi K3.1 Pro: 86.3%
Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis
- Fugu Max
- Kimi K3.1 Pro
| Leaderboard | Fugu Max | Kimi K3.1 Pro |
|---|---|---|
| Combined | Not in top 20 | #8 · 19.9% |
Ranks among Noumen catalog models on Combined leaderboards. Only top-20 placements are listed.
- Fugu Max
- Kimi K3.1 Pro
| Leaderboard | Fugu Max | Kimi K3.1 Pro |
|---|---|---|
| WebDev Overall | Not in top 20 | #6 · 1691.6 |
| WebDev · Webdev | Not in top 20 | #6 · 1691.6 |
| WebDev · Webdev-html | Not in top 20 | #5 · 1702.2 |
| WebDev · Webdev-react | Not in top 20 | #7 · 1697.1 |
Ranks among Noumen catalog models on LMArena leaderboards. Only top-20 placements are listed.LMArena
- Fugu Max
- Kimi K3.1 Pro
| Leaderboard | Fugu Max | Kimi K3.1 Pro |
|---|---|---|
| Game Development | Not in top 20 | #6 · 1398 |
| Website | Not in top 20 | #6 · 1326 |
| UI Component | Not in top 20 | #1 · 1400 |
| Data Visualization | Not in top 20 | #12 · 1310 |
| SVG | Not in top 20 | #2 · 1327 |
| 3D | Not in top 20 | #2 · 1469 |
Ranks among Noumen catalog models on Design Arena leaderboards. Only top-20 placements are listed.Design Arena
Live routing data
Performance and reliability
Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.
- Request success (24h)
- 100.0%
- Reliable
- Healthy (98% or more)
- Degraded (90–98%)
- Down (under 90%)
- No requests
Developers
Use it through the API
Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.
curl https://www.noumen.xyz/api/v1/chat/completions \
-H "Authorization: Bearer $NOUMEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "sakana/fugu-max",
"reasoning_effort": "high",
"messages": [{ "role": "user", "content": "Explain this stack trace." }]
}'import os
from openai import OpenAI
client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
model="sakana/fugu-max",
messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)sakana
Related models
Other sakana models, starting with the same family.