It does make the token price for kv cache writes significantly cheaper (2x -> 1.25x), but it will only bring actual savings for shorter and more focused sessions. Perfect for subagents. Generally good for powerusers and super bad for vibecoders.

I’ll be switching from Opus to Sonnet for the chat and tweak more of my subagents to use Opus. Also, keep in mind that Sonnet has a separate weekly rolling wind from the Opus, so it might be really beneficial to make that move and keep haikus for algorithmic workflows and opus for subagents.