The gap between open and closed AI models is basically gone. The AI Conference i

By Leosgovio · October 2, 2026 · Curated by George's Blog

The gap between open and closed AI models is basically gone.

The AI Conference in San Francisco just wrapped up.

While most presentations were around architecture and harness, one takeaway kept coming up in almost every conversation.

Open models like GLM and Kimi are now close enough to the closed frontier that the gap barely matters on real work.

For coding, problem solving and day to day tasks the difference is small.

And the setups that actually work best are not one giant model but a mix of models, each doing the part it is good at.

The split is the interesting part. US labs are shipping closed while Chinese labs are shipping open.

Same performance band, opposite approach.

Most people I talked to think the industry moves toward open from here, hoping that American companies will eventually release more of it too.

That feels like the real shift. Not which model is slightly ahead this month but who is willing to put the weights out.