sakana

Sakana: Fugu Ultra

Sakana: Fugu Ultra is a paid text-and-image model from Sakana with a 1,000,000-token context window and a 128,000-token output ceiling. It supports tool use and reasoning, which makes it usable for agentic workflows and multi-step tasks. Structured output support is unconfirmed, so developers who depend on guaranteed JSON schemas should verify that before committing. The comparison problem here is straightforward: at $5.00 per million input tokens and $30.00 per million output tokens, Fugu Ultra sits at a premium price point, yet it carries no independent benchmark coverage at this time. Buyers cannot weigh that cost against verified performance scores. Teams exploring very long context processing or reasoning-heavy pipelines may still find it worth piloting, but anyone budgeting carefully or comparing against established models should treat its quality as unproven until third-party evaluations are available.

Quality Score
87/100
price + capability + benchmarks
Input Price
$5.00
per 1M tokens
Output Price
$30.00
per 1M tokens
Context Window
1,000,000
tokens
Model ID
sakana/fugu-ultra
Vendor
sakana
Released
June 2026
Tokenizer
Other
Input Modalities
text, image
Output Modalities
text
Max Output
128,000 tokens
Tool Calling
✓ supported
Structured Output
✓ supported
Reasoning Mode
✓ supported
Vision
✓ accepts images
Audio
no
Moderated
no

What it costs in practice

Computed from the current $5.00/M input and $30.00/M output rates. Run your own numbers →

JobTokensCost
Summarize a 50-page report 30k in / 1.5k out $0.20
Classify 1,000 customer emails 500k in / 50k out $4.00
A month of a busy support chatbot 5M in / 2M out $85.00

Quick answers

How much does Sakana: Fugu Ultra cost?
$5.00 per million input tokens and $30.00 per million output tokens.
What is Sakana: Fugu Ultra's context window?
1,000,000 tokens, roughly 1,500 pages of text in a single request.
Does Sakana: Fugu Ultra support tool calling?
It supports tool calling, structured output, a reasoning mode.
Can Sakana: Fugu Ultra process images?
Yes, it accepts image input.