The right model for the right task

"I always use the biggest model, just to be safe." I hear that a lot. It's a bit like taking the moving van every time you pop out to the bakery. You get there, but it takes longer and costs more.

Claude Code lets you choose the model and the thinking depth deliberately.

🔀 Switching models

• /model in the session lists the available models to choose from.
• claude --model sonnet sets it right at startup.
• The short names: opus for the hard cases, sonnet for everyday work, haiku for quick, simple tasks.

🧠 My favourite: opusplan

With /model opusplan, Opus does the planning in plan mode (part 10), and Sonnet implements the approved plan. The big model thinks, the faster model types. For larger features, that's our best mix of quality and usage.

🎚️ Setting the thinking depth

/effort sets how thoroughly Claude thinks, from low to max. For a single tricky question, just write "ultrathink" in the prompt. Claude then thinks especially hard for that one answer.

A word of caution:

Which models and levels you see depends on your plan and on the model. And more thinking isn't automatically better: for simple tasks, max mainly means longer waits and higher usage. A look at /usage after a week shows what your choice really costs.

My rule of thumb:

• Renaming, adding tests, boilerplate? Sonnet or Haiku.
• Architecture, a nasty bug, unfamiliar code? Opus or higher effort.
• Bigger feature with a plan? opusplan.