Pick the Right Claude Code Model for Every Task
Most Claude Code users on Max or Team plans run Opus for everything — including tasks that Haiku handles in a third of the time at a fraction of the cost. With Opus 4.6 and Haiku pricing sitting at roughly a 15x difference, that habit is quietly expensive. A practical guide from Klement Gunndu on DEV Community breaks down five concrete patterns for matching model tier to task type, and the core insight is simple: the right model depends on what you're doing at that exact moment, not what you chose when you started the session.
The most immediately useful pattern is mid-session switching with /model — plan in Opus, implement in Sonnet, quick lookups in Haiku, no restart required. Close behind it is the opusplan alias, which automates that routing: Claude upgrades to Opus when you enter /plan mode and drops back to Sonnet when execution begins. For multi-agent workflows, the per-subagent model override is where the real cost leverage lives — spawning eight Haiku agents for parallelizable file search instead of eight Opus agents can cut per-session cost by 90% on the steps that don't require deep reasoning. The remaining two patterns cover CI/CD pipelines (use ANTHROPIC_MODEL to pin Sonnet in automation so you never accidentally trigger Opus pricing) and the priority chain for persistent defaults, where /model overrides --model overrides the env var overrides ~/.claude/settings.json.
The opusplan alias gets particular attention here as a genuinely underused feature — this is the first piece that explains exactly when the automatic model switch fires (plan mode entry and exit), which has been unclear since the feature shipped. If your team is running Claude Code at any kind of volume, building these patterns into your workflow is worth the twenty minutes it takes to read and configure.