This round of Claude updates doesn’t focus on “chatting better,” but instead pushes Claude a major step toward executable, sustainable workflows: the model adds an extended thinking mode; the API fills in the Files API and prompt caching; and the developer console makes cost tracking easier. Below, I’ll break down Claude’s new features by use case and explain them clearly.
Model-level update: extended thinking makes reasoning more reliable
Claude’s next-generation models offer two modes: instant replies, and extended thinking for deeper reasoning. For complex problems, long-chain decisions, or tasks that require repeated validation, Claude is more inclined to carry the reasoning through fully rather than just give a conclusion that “seems right.”
If you often use Claude for code design, debugging, or solution reviews, the value of extended thinking will be more evident: it’s better at clearly laying out assumptions, boundary conditions, and verifiable steps, reducing the back-and-forth needed to fill in missing information.
Agent capability upgrade: better endurance for long-running tasks
Claude’s agent capabilities have been enhanced, making it more friendly for “keep-running” task scenarios: it can maintain a to-do list within long-running workflows and, when granted local file access permissions, preserve continuity by writing to memory files. The officially disclosed capability limit supports independent operation on the order of several hours.
At the same time, Claude supports parallel handling of multiple tool calls, making tasks that require “look up information + edit files + run steps” more fluid. For teams, upgrades like these are more practical than incremental improvements in single-answer quality.
