01What is Sakana Fugu Ultra 1.1?

Sakana Fugu Ultra 1.1 is a new AI model from Japan that uses a council of experts to answer your prompt. Instead of one brain doing all the work, it sends your task to up to three different AI models, then merges their answers into one final output. This makes it more powerful than single models like Fable 5.

In this episode, Julian Goldie showed how it works with game builds. He gave it one prompt, and it created complete 3D games, open worlds, and flight simulators—all on the first try. It matched the best model on the planet in his tests.

02How the Council of Models Works

The core idea is simple: you give one prompt to Sakana Fugu, and an orchestrator routes it to one to three experts. Each expert works separately on part of the task, then their answers are merged into one final build. This is different from using a single AI brain.

Julian calls this the “one brain ceiling” problem. When one brain does everything—planning physics, painting graphics, wiring scoreboards—it can drop details when the job gets big. A council of models covers each other's mistakes, so you get better results.

  • One prompt goes to the orchestrator.
  • The orchestrator picks up to three experts.
  • Experts work independently.
  • Their answers are fused into one final answer.

03Setting Up Sakana Fugu Ultra 1.1

To start using Sakana Fugu Ultra 1.1, you have two main options. First, go to console.sakana.ai to get access directly. If that doesn't work because of your region, use OpenRouter instead.

Julian showed that it may be blocked in the UK and EU, so use OpenRouter there.

  1. Go to console.sakana.ai or OpenRouter.
  2. Create an account and get an API key.
  3. If you use the AgentOS, plug the API key into the Sakana Fugu section.
  4. Give a detailed prompt for your task.
  5. Wait about 25 minutes for the orchestrated thinking.
  6. Review the output in your workspace.

04Using Sakana Fugu for Game Builds

Julian used Sakana Fugu Ultra 1.1 to create several games on GoldieBench. He made a 3D game that was fun to play, an open-world game like Skyrim, and a flight simulator. All of them worked first time, with no back-and-forth needed.

He explained that testing with games is a good way to see how well the model plans, handles logic, codes, and designs UI. When you use it, you give one prompt, and the system builds everything for you. You can also use it for business automations, but games are a better test of its abilities.

  • Give a clear, detailed prompt describing what you want.
  • Let the system run for 15-25 minutes.
  • Check the output in your workspace.
  • If the result is good, keep it for reuse.

05Results and Comparison with Fable 5

Julian compared Sakana Fugu Ultra 1.1 with Fable 5 on GoldieBench. For the same game prompt, Fable 5 was a bit laggy and didn't work well with controls. Sakana Fugu Ultra 1.1 created a better, less linear version with better graphics, smoother performance, and fewer bugs.

In the flight simulator test, Fable 5 had less detail and wasn't as interesting. Sakana Fugu Ultra 1.1 was more fluid, more fun to use, and had better controls. Julian said it beat every other model he tested on these benchmarks.

  • Better graphics
  • Smoother performance
  • Easier controls
  • Less buggy

06Tips, Limits, and Mistakes to Avoid

One big limit is that Sakana Fugu takes about 25 minutes per shot, and you can't go back and forth. So don't use it for simple tasks like coding a basic website. Save it for big, complex jobs where you need frontier-level quality.

Also, be aware that it may be blocked in the UK and EU. If you're in Asia or the US, it usually works. As alternatives, you can use Fusion from OpenRouter or Hermes mixture of experts, which work in a similar way.

  • Use it for big jobs, not small ones.
  • Run it in the background to save time.
  • Test it yourself instead of trusting benchmarks alone.
  • Don't stick to one model—route jobs to the best fit.

07Why Sakana Fugu Matters for AI Users

Sakana Fugu Ultra 1.1 matters because it shows that combining multiple AI models can beat single frontier models. Julian says people winning with AI run a stack like the AgentOS and switch between models based on the job.

Even though slow models might seem not worth it, running them in the background while you do other work works well. If you need one-off super frontier outputs, the cost is worth it.

08Where to Get Sakana Fugu and AgentOS

You can get Sakana Fugu Ultra 1.1 from console.sakana.ai or OpenRouter. If you want the full setup, Julian offers the AgentOS inside his AI Profitable Boarding community. It includes memory, orchestration tools, and Sakana Fugu plugged in.

Inside the community, you get video tutorials, daily updates, weekly coaching calls, and a map to meet other AI builders. If you're non-technical, it's designed to be easy—over 205 pages of testimonials show wins.