We just added thinking block preservation in the Claude API. You can now control how thinking blocks are managed in your context window, resulting in more cache hits and lower costs.

→ View original post on X — @alexalbert__,