🎙️ EP 52: The Model That Took the “Open” Out of OpenAI (And It’s Free to Run)

What if the most powerful open AI model today wasn't from OpenAI?In this episode, we break down how Moonshot AI’s Kimi K2 just dropped a 1T-parameter beast that outperforms GPT-4.1 in real coding tasks—and anyone can run it without API limits.

We’ll talk about:

• The open-weight model crushing SWE-bench with 65.8% (better than GPT-4.1!)
• How Moonshot solved a major training bottleneck with MuonClip
• Why Meta is quietly closing off its "open" AI future
• Elon Musk’s pivot from hate-speech Grok to anime waifus (yep, really)

Keywords: Moonshot AI, Kimi K2, GPT-4.1, Claude Opus, open-source AI, SWE-bench, Meta AI, Grok, xAI

Links:

1. Newsletter: