Updates include: cache-aware rate limits, simpler prompt caching, and token-efficient tool use. Implement with minimal code changes:

→ View original post on X — @anthropicai