MODEL DIRECTORY
Sakana AI Model Directory
Sakana AI develops AI models and technologies inspired by nature and evolutionary principles.
WebsitePublic models
5
With relay prices
0
Relay sites
0
Reasoning
5
Tool call
5
Structured output
0
Multimodal input
5
Text models
5
Image models
0
Video models
0
Audio models
0
Release range
2026-06-15 ~ 2026-09-11
5 models foundPrices in USD per 1M tokens; relay prices and ≈ are converted from CNY at today's rate
| Model | Provider | Type | Relays | Context | Max output | Input | Output | Capabilities | Price (input / output) | Details | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
Fugu Maxfugu-max | Sakana AI | Text | - | 1M | 128K | Text: SupportedImage: SupportedPDF: SupportedFile: Supported | Text: Supported | Reason: SupportedTools: SupportedJSON: UnknownTemperature: Not supported | - | OfficialNo priceLowestNo price | 2026-09-11 | |
Multi-agent model for routing expert agents across complex analytical tasks Basics
Limits
InputText: SupportedImage: SupportedAudio: Not supportedVideo: Not supportedPDF: SupportedFile: Supported OutputText: SupportedImage: Not supportedAudio: Not supportedVideo: Not supported API capabilitiesReasoning: SupportedTool call: SupportedStructured output: UnknownTemperature: Not supported Pricing
| ||||||||||||
Fugu Ultra v2fugu-ultra-v2 | Sakana AI | Text | - | 1M | 128K | Text: SupportedImage: SupportedPDF: SupportedFile: Supported | Text: Supported | Reason: SupportedTools: SupportedJSON: UnknownTemperature: Not supported | - | OfficialNo priceLowestNo price | 2026-09-11 | |
Quality-first multi-agent model for hard research, analysis, and competitions Basics
Limits
InputText: SupportedImage: SupportedAudio: Not supportedVideo: Not supportedPDF: SupportedFile: Supported OutputText: SupportedImage: Not supportedAudio: Not supportedVideo: Not supported API capabilitiesReasoning: SupportedTool call: SupportedStructured output: UnknownTemperature: Not supported Pricing
| ||||||||||||
Sakana Namazusakana-namazu | Sakana AI | Text | - | 262.1K | 65.5K | Text: SupportedImage: SupportedPDF: SupportedFile: Supported | Text: Supported | Reason: SupportedTools: SupportedJSON: UnknownTemperature: Supported | - | Official$0.95 / $4LowestNo price | 2026-08-03 | |
Japanese-specialized reasoning model based on Kimi K2.6 and tuned for Japanese language, culture, and business workflows Basics
Limits
InputText: SupportedImage: SupportedAudio: Not supportedVideo: Not supportedPDF: SupportedFile: Supported OutputText: SupportedImage: Not supportedAudio: Not supportedVideo: Not supported API capabilitiesReasoning: SupportedTool call: SupportedStructured output: UnknownTemperature: Supported Pricing
| ||||||||||||
Fugufugu | Sakana AI | Text | - | 1M | - | Text: SupportedImage: SupportedFile: Supported | Text: Supported | Reason: SupportedTools: SupportedJSON: UnknownTemperature: Not supported | - | OfficialNo priceLowestNo price | 2026-06-15 | |
Multi-agent model for routing expert agents across complex analytical tasks Basics
Limits
InputText: SupportedImage: SupportedAudio: Not supportedVideo: Not supportedPDF: Not supportedFile: Supported OutputText: SupportedImage: Not supportedAudio: Not supportedVideo: Not supported API capabilitiesReasoning: SupportedTool call: SupportedStructured output: UnknownTemperature: Not supported Pricing
| ||||||||||||
Fugu Ultrafugu-ultra | Sakana AI | Text | - | 1M | - | Text: SupportedImage: SupportedFile: Supported | Text: Supported | Reason: SupportedTools: SupportedJSON: UnknownTemperature: Not supported | - | Official$5 / $30LowestNo price | 2026-06-15 | |
Quality-first multi-agent model for hard research, analysis, and competitions Basics
Limits
InputText: SupportedImage: SupportedAudio: Not supportedVideo: Not supportedPDF: Not supportedFile: Supported OutputText: SupportedImage: Not supportedAudio: Not supportedVideo: Not supported API capabilitiesReasoning: SupportedTool call: SupportedStructured output: UnknownTemperature: Not supported Pricing
| ||||||||||||
Capabilities and limits come from public sources. “Unknown” means no reliable data yet, not “not supported”. Relay counts include only sites with a current public price, de-duplicated per site.
Data source: models.dev · OpenRouter
Intelligence scores come from the OpenRouter benchmark rankings (Intelligence Index); higher is smarter.