Match the model to the job. Learn when to reach for Opus, Sonnet, or Haiku, what Opusplan does, and how to ask Claude to think harder on tricky problems.
Why: the models trade intelligence for speed and cost, so the right pick depends on the task. When: switch per task with /model, or set a default. Where: Sonnet is the balanced default for coding; reach up to Opus for the hardest work and down to Haiku for fast, simple tasks.
Opus — Most capable — best for deep reasoning, architecture, and long agentic runs. Slower and pricier.Sonnet — The balanced default — strong at everyday coding with good speed and cost.Haiku — Fastest and cheapest — great for simple, well-scoped tasks.Opusplan — Uses Opus to plan and Sonnet to execute — top-tier planning at a lower running cost./model # interactive picker
/model opus # switch now
/model opusplan # Opus for planning, Sonnet for the doingWhy: rather than switching every session, pin a default that fits your work. When: set it in settings.json for a project, or via an environment variable for all sessions. Where: a per-project default lives in .claude/settings.json so your team shares it.
// .claude/settings.json
{
"model": "opusplan"
}Why: extra reasoning helps on tricky logic, tricky bugs, and design — and you can ask for it in plain words. When: add a "think" cue for harder problems; escalate the phrase for more depth. Where: it costs more tokens, so save it for problems that actually need it.
Escalate reasoning right in your prompt:
"think" — a bit more reasoning
"think hard" — more
"ultrathink" — maximum reasoning budget
Example:
> ultrathink about why this race condition only happens under load,
then propose a fix.