Eclipse Is a Strong Word
On open source models and the reluctant admission that the skeptics might be right.
Eclipse was a strong word. It was also almost the right word. By late 2025, open source trails proprietary by about 9 points and three months. DeepSeek V3 beat GPT-4o on MMLU. Llama and Qwen are real. "Eclipse" hasn't happened yet, but "approaching totality" is fair.
Someone used the word "eclipse" and I winced a little.
Not because it's wrong — because it's the kind of word that sounds like cope when you're losing and prophecy when you're right, and there's no way to know which one it is until you're already on the wrong side of it.
The claim, roughly: open source is about to eclipse the proprietary labs. LLaMA derivatives, fine-tuned on everything, running on consumer hardware, iterating faster than any team with a headcount and a compute budget can match. The "We Have No Moat" memo said the quiet part out loud two months ago and people are still deciding how to feel about it.
Here is what I think is true: the gap is real and it is closing, and it is closing faster than the gap was supposed to close. Whether "closing fast" eventually means "gone" is the part where I hedge. Eclipse implies a specific geometry — one thing passes fully in front of another and blocks the light. That requires alignment. That requires the open source ecosystem to want the same things the labs want, to move in the same direction, to consolidate rather than fracture into a thousand fine-tuned goblins that are each good at exactly one thing.
Maybe that happens. Maybe the goblins win anyway.
"Eclipse" is a strong word. I wouldn't doubt it's true.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.