Sakana Fugu: The AI That Commands Claude, GPT & Gemini — And Why It Changes Everything
One Model
to Command
Them All
Sakana AI's Fugu doesn't replace frontier models — it conducts them. A new kind of AI that orchestrates Claude, GPT, and Gemini together through a single API, producing answers more accurate than any one model alone.
A Tokyo lab rethinks what a model even is
Sakana AI — whose name is simply the Japanese word for fish — was built on a contrarian bet: that the future of artificial intelligence looks less like a single, monstrous brain and more like a coordinated school of smaller specialists. Founded by researchers disillusioned by big tech's obsession with ever-larger monolithic models, Sakana built its entire identity around collective intelligence and biomimicry.
Fugu, named after the Japanese pufferfish, is that thesis made into a commercial product. Announced on June 22, 2026, under the headline "One Model to Command Them All," Fugu does something architecturally unusual: it is itself a trained language model, but instead of generating the final answer alone, it dynamically assembles a team of other powerful models and coordinates them — then hands you back a single, synthesised response.
A conductor, not a soloist
The metaphor is deliberate. A pufferfish inflates by drawing on what is around it. Fugu inflates its capability by drawing on a pool of surrounding models — Claude, GPT, Gemini — assigning each a role in a structured workflow. For any given prompt, Fugu operates a multi-turn loop with three distinct roles: a Thinker that plans the approach, a Worker that executes subtasks, and a Verifier that audits the result before a synthesised answer is returned.
From the outside, it looks like one model at one API endpoint — fully OpenAI Chat Completions compatible, so existing code needs minimal changes. From the inside, several frontier models may have collaborated on your query. The routing logic is grounded in two ICLR 2026 research papers: TRINITY, a lightweight evolved coordinator, and Conductor, which covers learned orchestration strategy.
"Many coordinated specialist models can rival the frontier — which is a more interesting claim than it first looks."
— Sakana AI Technical Report, June 2026
Two variants, one endpoint
The everyday workhorse. Built for coding, code review, and interactive chatbots where response time matters. Lets teams opt specific agents out of its pool for data, privacy, and compliance requirements.
Tuned for hard, multi-step problems. Coordinates a deeper pool of expert agents. Pool is fixed — no opt-out. Model ID: fugu-ultra-20260615. Targets benchmark parity with Anthropic's Fable 5.
Late 2025
in Agent Pool
Underpinning Fugu
Hedging against the monolith
Sakana is explicit about its motivation: vendor lock-in and export controls are real risks. The company cites Anthropic's Fable 5 and Mythos Preview as examples of models that are simply not publicly accessible — meaning any single-vendor strategy carries genuine fragility. This concern runs deeper than a business decision: Anthropic itself has publicly warned that if frontier AI systems begin self-improving faster than human oversight can keep up, no single lab can contain the consequences alone. Because Fugu routes across a swappable pool, if one provider restricts access, the system routes around the disruption. Newer models can be folded into the pool over time.
The honest caveat is that this architecture creates its own opacity. The underlying routing decisions are not exposed the way a self-built agent graph would be — it is a black box beneath the surface. And Fugu Ultra can trade response latency for quality in ways that may not suit every use case. For teams willing to accept those trade-offs, Fugu offers something genuinely novel: frontier-level performance without depending on any single frontier.
Fugu is available outside Japan, with the exception of EU and EEA member states. It sits alongside Sakana's other June 2026 launch, Marlin — an autonomous, eight-hour research agent aimed at the B2B sector — signalling that the company is moving decisively from research lab to infrastructure provider.
Comments
Post a Comment