Sakana says Fugu Ultra v1.1 router now tops Fable 5
Sakana AI's upgraded Fugu Ultra v1.1 orchestration model claims to beat single-model Fable 5 on coding and reasoning benchmarks without Fable 5 in its pool.
The claim
Sakana AI says its updated model router, Fugu Ultra v1.1, now outperforms Anthropic's standalone Fable 5 on coding and reasoning benchmarks — even though Fable 5 is not among the models it routes between. The Tokyo-based lab reported gains of up to 7.9 points over the prior v1.0, with the strongest improvements on ProgramBench and Terminal Bench 2.1, attributing the jump to incorporating newer frontier models into its pool.
How Fugu works
Fugu is an orchestration system rather than a single model: given a task, it decomposes and routes work across a pool of frontier LLMs, then synthesizes the results, exposed through an OpenAI-compatible API. That makes the comparison an apples-to-oranges one — a multi-model pipeline versus one monolithic frontier model — a caveat Sakana and observers acknowledge. Fugu launched in June 2026, days after Fable 5 was pulled from public access under US export controls, and its openly available positioning has been part of the appeal.
The significance for architecture
The result advances a live debate over whether orchestration and test-time synthesis can substitute for ever-larger single models. If a router over accessible models can match a restricted frontier system, the practical value of any one model's exclusivity weakens — and buyers gain a path to frontier-class output without direct access to the top model.
Why it matters
Sakana's pitch matters most where Fable 5 is unavailable. For developers shut out by export controls or price, an orchestration layer that reaches comparable quality reframes the market: the unit of competition becomes the routing system and its underlying pool, not a single flagship. Benchmarks from a vendor warrant caution, but the architectural argument — that coordination can rival scale — is gaining traction across the industry.