Claude Code v2.1.174: Model Picker Reliability Overhaul

Five fixes to the /model picker

Claude Code v2.1.174 focuses on the reliability of the /model picker, resolving a cluster of bugs that had been confusing subscribers across plan tiers.

What was fixed

• Opus disappearing for Max and Enterprise subscribers — Opus was incorrectly being hidden from the picker for some Max and Enterprise users who were entitled to it. It now appears as expected.
• Incorrect Sonnet version labels — when an environment variable pinned a specific Sonnet version, the picker displayed the wrong version label. The label now matches the pinned version.
• False "Fable 5 consuming usage credits" banner — enterprise accounts saw a misleading banner suggesting Fable 5 was drawing down usage credits when it was not. The banner has been removed for those accounts.
• Advisor model pre-selecting blocked models — the advisor flow could pre-select a model the user was not permitted to use; it now respects model availability.
• Background sessions inheriting the parent shell's provider environment — background sessions picked up the provider environment of the shell that launched them, leading to inconsistent model routing. Background sessions are now isolated from that inheritance.

Why it matters

The model picker is the control surface every Claude Code user touches to choose how their work is executed. When it shows the wrong models, mislabels versions, or surfaces inaccurate billing warnings, it undermines trust and can route work to the wrong model. This release cleans up the most visible of those papercuts, particularly for Max and Enterprise users who were most affected.