Sakana AI didn't build a bigger model. It built a switchboard, and called it a model anyway. Fugu, released by the Tokyo lab on June 22, 2026, doesn't compete with Claude, GPT, or Gemini so much as it sits on top of all three, deciding which one should answer your question. That framing alone made it one of the most argued-about AI releases of the summer.
What Fugu actually is
Sakana AI is not a newcomer improvising a marketing stunt. The three-year-old startup was founded by CEO David Ha, CTO Llion Jones — one of the eight co-authors of the original "Attention Is All You Need" Transformer paper — and COO Ren Ito, all ex-Google researchers. Fugu is their answer to a question the rest of the industry has been dodging: what happens when no single model is reliably available?
Fugu ships in two versions. Standard Fugu handles everyday, latency-sensitive requests: drafting an email, debugging a function, answering a question. Fugu Ultra is built for slower, harder problems: multi-step research, cybersecurity analysis, patent review. Neither is a conventional LLM in the sense of one set of weights answering everything. Both are themselves language models trained to receive a task, decide whether to handle it directly or hand pieces of it to specialist models in an agent pool — Gemini, DeepSeek, Kimi, and, when available, Claude and GPT — verify what comes back, and stitch the results into one response. It can even delegate to other instances of itself.
Why this, why now
The timing is the real story. On June 12, 2026, the U.S. Department of Commerce placed export controls on Anthropic's two most capable models, Claude Mythos Preview and Claude Fable 5, cutting off access outside the U.S. essentially overnight. For enterprise teams outside American jurisdiction who'd built workflows around those models, that was a supply-chain problem, not an abstract policy story. Sakana CEO David Ha positioned Fugu explicitly as the fix: route around any one vendor's outage, ban, or price hike by keeping several providers in the pool at once.
TechCrunch clocked it as part of a pattern, not a one-off: Asian AI labs launching Mythos-adjacent systems while Anthropic's export ban drags on. Fugu is the most credentialed entrant so far, but it won't be the last.
The benchmarks say one thing, the users say another
On paper, Fugu Ultra looks formidable. It posts 73.7% on SWE-Bench Pro, ahead of Claude Opus 4.8's 69.2%, GPT-5.5's 58.6%, and Gemini 3.1 Pro's 54.2%. On LiveCodeBench it scores 93.2 against Fable 5's 89.8. It reportedly beats the older Mythos Preview on GPQA-D, the graduate-level science benchmark.
Here's the catch worth stating plainly: none of those Fable or Mythos comparisons are head-to-head. Both models were pulled from public access by the same export order that inspired Fugu's existence, so Sakana benchmarked against Anthropic's own published reference scores, not a live run. On SWE-Bench Pro specifically, Fugu Ultra actually trails Fable 5's reported 80.0 — a number Sakana's own materials disclose.
AI researcher Ethan Mollick ran his usual coding tests against Fugu Ultra and found it "incredibly slow," with a single run taking 30 minutes. His verdict: results were fine, but fell short of Fable in practice.
Hacker News threads on the release were blunter. Commenters flagged that the $200-a-month Max plan burns through its allotment in under three hours a week of real use, that the API lags, and that output quality doesn't match what Fable 5 delivered before the export order. The pricing, for reference:
- API: $5 per million input tokens, $30 per million output tokens (Fugu Ultra)
- Standard subscription: $20/month
- Pro subscription: $100/month, 10x usage
- Max subscription: $200/month, 20x usage
Some estimates put Fugu Ultra's effective cost at five times Claude Opus 4.8's, for a result multiple testers describe as slower and, in practice, weaker.
The real argument
I'd argue the interesting critique isn't speed or price — it's the premise. Fugu doesn't eliminate dependency on frontier labs; it spreads that dependency across five of them and hides the seams behind one endpoint. That's a genuinely useful abstraction for a team that got burned when Fable 5 vanished from their stack overnight. It is not the same thing as building a sovereign model that doesn't need Claude, GPT, or Gemini to function, and Sakana has never claimed otherwise, whatever the headlines implied.
What Fugu proves is narrower and more useful than "Japan built a frontier model." It proves that orchestration — knowing which model to call, when, and how to reconcile the answers — is now a product category on its own, and that the export-control shock on June 12 created real demand for exactly that. Whether Fugu is the right implementation of that idea is still an open argument on Hacker News. That Sakana correctly identified the gap is not.
