- Big Players
- Posts
- The secret 1M token AI model that just appeared on the internet
The secret 1M token AI model that just appeared on the internet
Forget Claude 3.7 - this unknown model is 4x faster... but do AI models even matter?
Matthew Berman
April 08, 2025
What if you could reverse-engineer your industry's best content, rebuild it in your voice, and automate it across channels—by next week?
Over the weekend, I went viral tweeting about a mysterious new foundation AI model—Quasar Alpha. No one knows who made it. No big announcement. Just... appeared.
Quasar Alpha is here, and it's the AI industry's best-kept secret.
A mysterious 1M token context model that beats Claude 3.7 Sonnet on benchmarks while running 4X faster.
But nobody knows which lab created it.
Here's what we know so far: 🧵
— Matthew Berman (@TheMattBerman)
10:31 PM • Apr 4, 2025
A sucker for mystery, I had to test it. With a 1M-token context window (it can "remember" ~750,000 words), I built an agent to scrape content from a brand and its competitors, analyze patterns, and recreate what works—fully in the brand’s tone.
And I did it all using nothing but the context window.
It worked great. Cool automated pipeline with a few lines of code and some prompts.
And honestly—none of that even matters.
Because the model is just a raw material. And raw materials—on their own—don’t build empires. The real scale comes when you turn that material into a system: logistics (agentic systems), design (application layer), sales (distribution), and data flows that are uniquely yours. That’s how you build a moat.
That’s how you create a durable, compounding advantage.
This isn’t about picking the "best model." It’s about building smart, context-driven AI systems that create moats around your business.
Let’s break it down.