Sakana AI logo

Sakana Fugu Ultra

Sakana AI · Released Jun 22, 2026

ProprietaryreasoningagentscodingCompare this model

$5

Input price / MTok

#26 of 27

$30

Output price / MTok

#26 of 27

1M

Context window

#18 of 28

95.5

GPQA Diamond

#1 of 19

93.2

LiveCodeBench

#2 of 8

Summary

Sakana Fugu Ultra is a proprietary language model from Sakana AI, released on Jun 22, 2026.

At $5 input and $30 output per million tokens, it ranks #26 of 27 for input price, placing it among the higher-priced models tracked here.

Its context window of 1M tokens ranks #18 of 28 for context depth in this dataset.

It targets reasoning, agents, coding workloads.

Orchestrator model trained to devise agentic scaffolds across a swappable pool of frontier LLMs; behaves as a single model behind one API (coordination is learned, not static bundling).

Benchmarks

BenchmarkScoreMax%RankSource
GPQA Diamond95.510095.5%#1/19Sakana AI (self-reported in technical re...
LiveCodeBench93.210093.2%#2/8Sakana AI (self-reported in technical re...

Similar Models

Compare all 6 models

Models with the most shared benchmark coverage and closest scores to Sakana Fugu Ultra.

Sakana Fugu Ultra
DeepSeekDeepSeek V4 Pro
DeepSeekDeepSeek V4 Flash
xAIGrok 3
GoogleGemini 2.5 Pro
AlibabaQwen3.7 Max
GPQA Diamond#1 of 19
Graduate-level science reasoning
LiveCodeBench#2 of 8
Live competitive programming

Compare Sakana Fugu Ultra side by side

Pre-filled with 5 similar models

Open comparison