01What Fugu vs Fusion Is and How These APIs Work

Fugu vs Fusion is a comparison between two new panel AI APIs: Sakana's Fugu Ultra/Mini and OpenRouter's Fusion. Both let you send one prompt and get a synthesized answer from multiple models working together. A judge model critiques and combines the outputs into a single result.

Fugu is built by Sakana in Japan and comes in two versions: Fugu Ultra (high quality) and Fugu Mini (faster, lower quality). Fusion is from OpenRouter and uses five models behind the scenes. Both claim to beat the now-unavailable Fable 5 on benchmarks.

You can access both through their APIs, but they work differently. Fugu offers a flat subscription plan, while Fusion charges per token. In this episode, Julian Goldie tested them side by side on visual tasks to see which one delivers better results.

02Main Features of Fugu and Fusion

Fugu Ultra and Fugu Mini are the two tiers from Sakana. They generate high-quality visuals from a single prompt. Fusion from OpenRouter also works as a panel engine, combining answers from multiple models.

  • Both are one-prompt systems: you give one instruction and wait for the output.
  • Fugu Ultra produces the best visual quality among the three.
  • Fusion is accessible via OpenRouter API and can be plugged into other tools.
  • Fugu Mini is faster but lower quality than Fugu Ultra.
  • Fugu has a flat subscription plan; Fusion charges by token usage.

03How to Get Started with Fugu or Fusion

To use either API, you need to sign up for access and get an API key. Then you can connect it to a tool like the agent operating system shown in the episode.

  1. For Fugu (Sakana): Go to the Sakana website, create an account, and subscribe to the flat plan. You get a set number of tokens per day.
  2. For Fusion: Sign up at OpenRouter, get an API key, and pay per token as you use it.
  3. Connect to a workspace: Both APIs can be plugged into the agent operating system, which gives you a chat interface and a workspace to see outputs.
  4. Send your first prompt: Open the chat box, type your request (e.g., “create a landing page”), and wait 5-10 minutes for the panel to generate a response.

04How Julian Tested Fugu vs Fusion on Real Tasks

Julian ran three visual tests to compare Fugu Ultra, Fusion, and Fugu Mini. He gave each API the same prompt and evaluated the output quality.

  • Landing page test: Fugu Ultra produced a very nice, well-designed page. Fusion's page was still good but not as polished. Fugu Mini came in third.
  • Raycaster maze test: Fugu Ultra again won with smooth, cool-looking output. Fusion was decent but not as impressive.
  • Living Spiral Galaxy simulation: Fugu Ultra created a beautiful, interactive galaxy you could zoom and drag. Fusion's version was less smooth and less realistic.

Overall, Fugu Ultra won all three tests. Fusion beat Fugu Mini on two of them, but the gap was smaller.

05Results and Comparisons from the Episode

In the direct comparison, Fugu Ultra clearly outperformed Fusion on visual quality. The galaxy simulation especially stood out as “the best living spiral galaxy simulation” Julian had ever seen from an AI.

Fusion did better than Fugu Mini, but the difference was less dramatic. Julian also compared Fugu Ultra to Opus 4.8 on a liquid-in-a-bowl simulation. Fugu Ultra's version looked much more realistic and polished.

On benchmarks, both Fugu and Fusion claim to beat Fable 5, but Fable 5 is no longer available for testing. So the visual tests serve as a proxy for quality.

06Tips, Limits and Mistakes to Avoid

There are important limits to know before using either API. Fugu has a token cap and generation limit. Julian hit a five-hour cooldown after a few generations. Fusion has fewer limits but costs per token.

  • Wait times: Both can take 5-10 minutes to respond. You cannot tweak the prompt in real time.
  • One-shot only: If the output is truncated or buggy, you have to start over. No back-and-forth.
  • Fugu Mini is inconsistent: Some outputs are buggy or lower quality than Fugu Ultra.
  • API availability: Fusion relies on OpenRouter. If it goes down, you can switch to another model. Fugu is standalone.
  • Don't expect real-time chat: These are not like Claude or Hermes. You send a prompt and wait for the panel to generate.

07Who Should Use Fugu vs Fusion and Why It Matters

If you need high-quality visuals from a single prompt, Fugu Ultra is the winner. It creates beautiful, smooth, interactive outputs that can beat older models like Opus 4.8. It is ideal for developers and creators who want a flat subscription and don't mind limits on daily usage.

Fusion is better for those who need more flexibility. It charges per token, so you only pay for what you use. It also integrates easily with tools like the agent operating system.

If you need to generate many outputs without waiting, Fusion may be more practical despite lower quality.

Both APIs matter because they show that panel AI can rival and surpass older benchmarks like Fable 5. As Julian noted, you can even switch between them if one goes down. This makes them valuable for building reliable AI-powered applications.