My experience is that the models are close enough that, with good harness and context engineering, the open weight models can be better for your daily workflows, than frontier without the special treatment. Basically every point of difference made in this article is suspect once you start adding the h/c engineering. Benchmarks are not test scores as much as sign posts.
I'm now hoping the Chinese take the lead. The US needs a sputnik like event to break us out of our hubris and funk.
Let's pace AI to stay ahead.
My experience is that the models are close enough that, with good harness and context engineering, the open weight models can be better for your daily workflows, than frontier without the special treatment. Basically every point of difference made in this article is suspect once you start adding the h/c engineering. Benchmarks are not test scores as much as sign posts.
I'm now hoping the Chinese take the lead. The US needs a sputnik like event to break us out of our hubris and funk.