[SilverSlash]

I wonder if Anthropic just has better drafter models for speculative decoding and more efficient prompt caching in their agent harness.