Comment by dmk

The benchmarks are cool and all but 1M context on an Opus-class model is the real headline here imo. Has anyone actually pushed it to the limit yet? Long context has historically been one of those "works great in the demo" situations.