← all entries
B-TIER news

Claude — I TESTED IT OUT IN MY RECENT VIDEO - it’s great, but I had to try...

gigaqian · 2026-04-27T14:24:02.000Z · in AI Saves, Vol. 7
Open post ↗

100k ft — What is this?

Claude — I TESTED IT OUT IN MY RECENT VIDEO - it’s great, but I had to try...
newsB-tier

50k ft — Why care?

Orchestration models are the next frontier in Al! This is according to the CEO of Sakana, David Ha. Sakana Fugu is a ... — @gigaqian; uses Claude; demonstrates API integration.
prompt engineeringvideo production

10k ft — Key takeaways

(no takeaways yet)

5k ft — Tools & techniques

Claude API integration

1k ft — Deep-dive / try it

Orchestration models are the next frontier in Al! This is according to the CEO of Sakana, David Ha. Sakana Fugu is a collective intelligence Al system that dynamically orchestrates the world’s best available Al models to tackle complex tasks. Let’s talk about what makes this different. It’s a big moment! #breakingnews #newai #futuretech #japan #gigaqian sina.growthtech   1w I TESTED IT OUT IN MY RECENT VIDEO - it’s great, but I had to try hard to get that output • NO WHERE NEAR THE FABLE MABLE
Transcript excerpt
As of today, we now have something better than Claude Fable-5, at least according to the Japanese company behind this AI model. But it's more than a model, and that comes with some caveats. But the founder of this company believes that this is the next frontier in the AI race. I'm Giga Chen, and today I'm going to tell you all about Sakana Hugu. Let's talk about who built this. David Ha is the CEO. He was a derivatives trader at Goldman Sachs. Then he was a research scientist at Google Brain. Then he was head of research at Stability AI. He co-founded Sakana with Cleon Jones. Yes, that Cleon Jones co-author of the paper, attention is all you need, and the one who coined the word, transformer. Well, I'd say he's got his attention now. This comes in two versions. There's Hugu and Hugu Ultra. And in some areas, these models do indeed beat Claude Fable. So it must be a massive model, right? No, very wrong. The proprietary model is only 7 billion parameters. So how can I answer your prompts on the same level as these frontier models? Well, it can't, and it never actually answers your prompt. It was trained based on the methodologies and frameworks laid out in the Trinity paper and the conductor paper. And you can look both of these up. It was published by Sakana's internal research team. Back to the point, this is an orchestration model. So yeah, it doesn't answer your questions. Hugu is a foreman. And it handles both open and closed models, allowing them to do what they do best,…

More from @gigaqian

Collection: AI Saves, Vol. 7