That said Sonnet 4.5 isn’t new and there have been loads of innovations recently.

Exciting to see open models nipping at the heels of the big end of town. Let’s see what shakes out over the coming days.

If you run SWE-bench evals, just make sure to use the most up-to-date code from our repo and the updated docker images

I don't doubt that it's an oversight, it does however say something about the researchers when they didn't look at a single output where they would have immediately caught this.

Claude spits that very regularly at the end of the answer, when it's clearly out of it's depth, and wants to steer discussion away from that blind-spot.