AI export controls and the need for AI autonomy will bring new prospects to startups to take a bite out of big AI companies.
A Japanese startup named Sakana has recently released Fugu, a new AI model that works with developers and enterprises. In contrast to depending on just one single big AI model, Fugu employs several AI agents provided by various sources and automatically picks the most suitable agent for any task.
This ensures that businesses don’t experience any issues like vendor lock-in, disruptions in the services, or any limitations on its usage imposed by the AI vendor. In case one AI provider goes offline due to any regulation or technical problem, Fugu automatically switches to the other model and keeps the process going.
Sakana claims that this multi-agent solution will help companies to have a safer option and a lot of flexibility, especially when countries start to impose stricter limitations on the export of advanced AI models.
Benchmarking Results
According to Sakana, the AI platform Fugu Ultra has obtained comparable performance compared to top-performing AI models such as Claude Fable 5 and Mythos 5 from independent tests.
On tests for coding performance on LiveCodeBench, Fugu scored 93.2 and Fugu scored 92.9, with better scores than Claude Fable 5.
On the advanced science knowledge GPQA-D benchmark, Fugu models have obtained higher scores than the former Claude Mythos Preview model.
This indicates that Fugu is capable of providing high-performance AI with a flexible multi-model design.
Price and Accessibility
There are two types of Fugu provided by Sakana. The basic version of Fugu is created for quick replies and common activities.
The second one, Fugu Ultra, is a more advanced product which is used for conducting complicated jobs such as AI research, cyber security analysis, and in-depth patent investigations. It uses the network of a bigger number of AI agents.Fugu is not an open source AI system.
