BREAKING
Sakana AI Releases Fugu Ultra
How Fugu Orchestrates Models
1Select
2Delegate
3Verify
4Synthesize
GPQAD Benchmark Scores
Fugu Ultra95.1
Gemini 3.194.4
Opus 4.692.7
GPT 5.490.9
0$/M
input tokens
0$/M
output tokens
0$/mo
Standard plan
Promise vs Skepticism
UpsidePros
Strong on complex workflows
Surfaced 20+ code issues
No export-control risk
ConcernsCons
Latency of tens of seconds
Unclear which model runs
Self-reported benchmarks
Calls for Independent Tests
AI NEWS BLITZ
Tokyo's Sakana AI has launched Fugu Ultra, claiming frontier-level performance through model orchestration.