ModelBench
Sakana logo
Models / Sakana

Fugu

sakana/fugu

Multi-agent model for routing expert agents across complex analytical tasks

Source-linkedfuguClosed weightsReleased 2026-06-15
Report data
Context window1M
Maximum outputNot documented
Provider offers1
Knowledge cutoffUndisclosed
Capabilities

What this model supports

ReasoningYes
Tool callingYes
Structured outputYes
AttachmentsYes
Temperature controlNo
Open weightsNo
Creator
Sakana
Family
fugu
Modalities
Image, Text
License
Not documented
Release date
2026-06-15
Model ID
sakana/fugu
Market availability

Provider pricing

Report price →
Providers offering Fugu
ProviderProvider model IDContextInput / 1MOutput / 1MCached / 1MFeaturesAll pricing
Sakana AIfugu1M———R T S
View
{}
Evidence

Published benchmark results

CTI-REALM vs top 2 models
Fugu Ultra
69.4
Fugu
67.5
CharXiv Reasoning vs top 5 models
Muse Spark 1.1
88.4
Fugu Ultra
86.6
Fugu
85.1
Gemini 3.5 Flash
84.2
Gemini 3.1 Pro Preview
83.3
GPQA Diamond vs top 7 models
Fugu
95.5
Fugu Ultra
95.5
GPT-5.6 Sol
94.6
GPT-5.4 Pro
94.4
Gemini 3.1 Pro Preview
94.3
Claude Opus 4.7
94.2
GPT-5.5
93.6
Humanity’s Last Exam vs top 2 models
Fugu Ultra
50.0
Fugu
47.2
LiveCodeBench vs top 3 models
Fugu Ultra
93.2
Fugu
92.9
Nemotron 3 Ultra 550B A55B
89.0
LiveCodeBench Pro vs top 2 models
Fugu Ultra
90.8
Fugu
87.8
Long Context Reasoning vs top 2 models
Fugu
74.7
Fugu Ultra
73.3
MRCRv2 vs top 2 models
Fugu Ultra
93.6
Fugu
86.6
SWE Bench Pro vs top 2 models
Fugu Ultra
73.7
Fugu
59.0
SciCode vs top 7 models
Fugu
60.1
Fugu Ultra
58.7
Qwen3.7 Max
53.5
Gemini 2.5 Pro
42.8
GPT-5-Codex
40.9
Step 3.5 Flash
40.4
Gemini 2.5 Flash
39.4
Terminal Bench 2.1 vs top 2 models
Fugu Ultra
82.1
Fugu
80.2
τ3 Banking vs top 2 models
Fugu
21.7
Fugu Ultra
20.6