For the past few years, the AI industry has been obsessed with building bigger, smarter and more capable models. Now, a new release from Sakana AI suggests the next AI arms race may not be about building the best model at all, but rather building the best system for managing multiple models.
As Anthropic's Fable 5 and Mythos models generate intense discussion across the industry about performance, capabilities and access, Sakana AI has introduced a new approach with Fugu.
The timing of this release is especially interesting as users are increasingly stacking models and utilizing several AI agents at once.
What is Sakana Fugu?
Unlike ChatGPT, Claude or Gemini, Fugu is not trying to be the smartest model in the room. Instead, it acts more like an AI project manager.
When a user submits a task, Fugu analyzes the request, decides which AI models are best suited for different parts of the problem, routes work to those models, evaluates the responses and combines the results into a final answer.
It's similar to a manager assembling a team of specialists instead of relying on a single employee. Because we all know one model isn't good at everything. Instead, one model might be better at coding, while another might excel at reasoning or writing. Simply put, Fugu's job is to determine who should do what and then stitch everything together.
According to Sakana AI's website, this orchestration approach allows the system to achieve performance comparable to leading frontier models without depending entirely on a single model provider.
Why this matters
Most people think of AI competition as a race to build the biggest and most powerful model, but Fugu points toward a different possibility. Instead of models attempting to outperform each other individually, what if the future belongs to systems that know how to combine multiple models effectively.
And while this concept is anything but new, what makes Sakana stand out is it has trained the orchestration process itself and made the routing intelligence the centerpiece of the product. In other words, it made the coordinator as important as the workers.
The lesson from Fable
The conversation surrounding Anthropic's Fable models highlighted something many organizations are beginning to recognize and that's relying on a single AI provider can create challenges. When access changes, outages occur, pricing shifts or capabilities evolve, entire workflows can be affected overnight.
Systems like Fugu are designed to reduce that dependency. Rather than building around one model, they build around an ecosystem of models. So now, if one model becomes unavailable, another can potentially take its place. If a better model emerges tomorrow, it can theoretically be added to the mix.
That flexibility could become increasingly valuable as the AI landscape grows more competitive.
The takeaway
Don't get me wrong: model size, benchmark scores and raw capability still matter. But Sakana's Fugu hints at a future where the most important question isn't "Which model is best?" but rather "Which system is best at choosing the right model?" Fugu suggests the next phase of competition may look very different. Instead of creating a single AI that does everything, the winners may be the companies that can assemble, coordinate and optimize entire teams of AIs behind the scenes.
If that's the direction the industry is heading, the next breakthrough might be an AI smart enough to know when not to answer the question itself.
Follow Amanda Caswell and stay ahead of the AI curve
