Post

The nuxt evals result is the shape of the argument now: good enough local models erode the moat from below. The frontier labs keep winning on the hardest 5% of tasks, but most of the money is in the boring 95%.