Summary & Key Takeaways ​

• Microsoft announced two new LLMs: MAI-Thinking-1 (35B) and MAI-Code-1-Flash (5B).
• MAI-Thinking-1 is claimed to outperform Sonnet 4.6 in blind evaluations.
• MAI-Code-1-Flash is purpose-built for GitHub Copilot and VS Code.
• Both models were trained on "enterprise grade, clean and commercially licensed data."
• The low parameter counts are notable given their claimed performance.

Our Commentary ​

Okay, this is big. Microsoft releasing smaller LLMs that claim to punch above their weight, and explicitly stating "commercially licensed data"? That's a direct shot at the current legal quagmire surrounding AI training data. If these models deliver on performance and licensing, it could fundamentally shift the landscape. I'm skeptical but hopeful. The "appropriately licensed" part is what I'm really watching.