Japanese AI startup Sakana AI has launched Fugu, a new system that takes a different approach to building a competitive AI model: instead of relying on one large model, it coordinates multiple AI models on the fly.
According to The Decoder, Fugu orchestrates several large language models together to match the benchmark performance of leading systems, specifically Anthropic's Fable 5 and Mythos. The Decoder reports that part of Sakana's goal is to reduce dependence on any single AI provider — a notable pitch at a time when companies worry about being locked into one vendor.
Moneycontrol describes Fugu as a new AI model family, and reports that Sakana claims it outperforms Anthropic's Claude Fable 5 on coding and scientific reasoning benchmarks. Both Moneycontrol headlines refer to Anthropic's Claude Fable 5 as "banned," though the source items provided here do not spell out the details of that restriction.
The underlying idea is sometimes called orchestration: rather than asking a single model to do everything, a controller routes work to whichever models are best suited to a given task. Sakana AI, known for research-driven and nature-inspired approaches, is positioning Fugu as a way to reach top-tier results by combining models rather than building one ever-larger system.
It's worth noting these are company claims tied to benchmark tests, and the sources here do not include independent verification of the results.
Why it matters: if coordinating several models can rival the best single models, it could give companies a cheaper, more flexible path to high-end AI — and reduce their reliance on any one provider.