Claude Code supports three model-switching methods: /model command (instant), --model flag (session), and shell env var (permanent default across ZSH/BASH).
Key Takeaways
/model command switches models inside an active Claude Code session with no terminal restart required.
--model flag scopes the model choice to one session; useful for one-off Opus runs without changing your default.
Permanent defaults go into .zshrc (macOS) or .bashrc (Linux) and require a full terminal restart and fresh claude invocation to take effect.
The help center article covers model availability in Claude Code specifically, not the broader API or Claude.ai web interface.
Hacker News Comment Review
An Anthropic engineer confirmed this was a stale support article written before Opus was added to Pro plans in January (Opus 4.5 rollout); the doc was never updated, making the HN headline misleading.
Commenters who tried Opus on Pro report it consumed quota so aggressively it was effectively locked to Max plan subscribers anyway, suggesting the restriction is practical before it is formal.
Broader thread concern: investor-subsidized inference is ending; open models are the only low-cost escape, but Chinese-origin weights create enterprise trust blockers that keep proprietary pricing sticky.
Notable Comments
@underyx: Pulled Wayback Machine evidence showing the support article predates January’s Opus 4.5 Pro rollout; Anthropic confirmed and apologized for the confusion.
@spudlyo: Catalogues a streak of Anthropic self-inflicted trust damage: muddled plan messaging, reasoning-effort changes, caching bugs, and corporate denials.
@thoughtlede: Argues investor capital may shift from funding model providers to funding enterprises that can prove AI ROI, reshaping who pays for Opus-level inference.
