Claude Code Now Lets You Switch Models Mid-Session

One command. Immediate effect. No terminal restart required.

Anthropic has updated Claude Code with three distinct methods for controlling which model powers your session. The /model command switches models interactively from inside an active session. The --model flag sets the model at launch for a single session. Environment variables lock in a permanent default across all future sessions. Supported models include Opus 4.7, Sonnet 4.6, Opus 4.6, Haiku 4.5, and Sonnet 4.5.

If you've been defaulting to the same model for every task, this update is a prompt to stop doing that.

Three Configuration Methods, Three Different Use Cases

The /model command is the right choice when you're mid-session and the task shifts. You're drafting copy with Sonnet and suddenly need to run a complex competitive analysis—type /model, select Opus from the interactive menu, and keep working. No restart, no friction. Run /status anytime to confirm which model is currently active.

The --model flag is for when you know going in that a session has specific requirements. Starting a session focused entirely on high-volume metadata generation? Launch with Haiku from the start:

claude --model claude-haiku-4-5-20251001

The flag applies only to that session. Your default is unchanged.

Environment variables are the permanent solution—the right move for anyone who's identified their preferred daily-driver model and doesn't want to configure it manually every time.

For zsh users on macOS: echo 'export ANTHROPIC_MODEL="claude-sonnet-4-6"' >> ~/.zshrc source ~/.zshrc

For bash users on Linux: echo 'export ANTHROPIC_MODEL="claude-sonnet-4-6"' >> ~/.bashrc source ~/.bashrc

Close your terminal completely, reopen, and your chosen model is the new default for every session going forward.

Which Model for Which Task

The configuration flexibility is only valuable if you have a working model for when to use each tier. Here's how we think about it for marketing and growth workflows:

Haiku 4.5 for high-volume, templatable tasks where speed and cost matter: metadata generation, first-pass tagging, structured data extraction, draft outlines. Don't burn Opus compute on work Haiku handles cleanly.

Sonnet 4.6 as the reliable daily driver. Brief writing, campaign analysis, content iteration, most client-facing copy. Capable enough that quality rarely suffers, efficient enough for sustained use across a full workday.

Opus 4.7 for the sessions that genuinely require depth: complex strategic analysis, nuanced long-form content, reasoning tasks where the quality gap between model tiers shows up visibly in the output. This is your senior resource—deploy it accordingly.

The principle is simple: match compute to complexity. Running every task through Opus is expensive and often unnecessary. Running everything through Haiku is penny-wise and output-poor. The new configuration options make it practical to actually maintain that discipline.

One Practical Setup Recommendation

If you haven't set an environment variable default yet, Sonnet 4.6 is the sensible starting point for most marketing and growth workflows. It covers the majority of daily use cases without requiring you to actively manage model selection session to session. Reserve the /model switch for the Opus sessions that earn it—and the Haiku sessions where you're deliberately optimizing for volume.

Five minutes of setup. Measurable difference in how efficiently your Claude Code sessions run.

The Winsome Marketing team works with growth engineers and technical marketers building exactly these kinds of optimized AI workflows. If you're thinking about how to structure Claude Code into your production stack, let's talk.