CloudCodeTree LogoCloudCodeTree
AI NewsTutorialsAbout
CloudCodeTree Logo
CloudCodeTree
  • AI News
  • Tutorials
  • About
← Back to AI News
Two Managed Settings Keep Your Whole Team on the Approved Model List in Claude Code

Two Managed Settings Keep Your Whole Team on the Approved Model List in Claude Code

Chris Harper

2 min read

Aug 26, 2026 · 04:08 UTC

AI
Workflow
Claude Code
Best Practices

TL;DR: modelPicker curates your team's /model picker to only approved models; modelPricing plugs in contracted rates so cost estimates match your invoices. Both are managed settings in v2.1.243 (August 25).

If you're running Claude Code across a team, you've probably hit this: developers on different models, cost reports that don't match invoices, and config that drifts as people try new releases.

Two managed settings added in v2.1.243 address this directly.

modelPicker replaces the default /model picker with your own ordered, labeled list — accepts any model ID including Vertex and Bedrock identifiers. You define what appears; the developer still chooses within that list.

"modelPicker": [
  { "id": "claude-sonnet-5-20260801", "label": "Team Standard" },
  { "id": "claude-opus-5-20260901", "label": "Deep Review (expensive)" }
]

Combine with ANTHROPIC_DEFAULT_MODEL (set in your org's managed settings) to ensure every new session starts on your preferred default without per-developer config.

modelPricing feeds in your org's contracted per-model rates and discount multiplier, so /cost, the status line, and usage telemetry show what you actually pay rather than Anthropic list price. Without it, every cost estimate is wrong for anyone on a negotiated contract — a real problem for teams tracking spend against budget.

The limit you should know: modelPicker controls the TUI menu, but it does not block someone from setting ANTHROPIC_MODEL directly in their shell. If you need model enforcement at the infrastructure level — for compliance or cost control — a gateway proxy that restricts model IDs at the API layer is the right control; this is a convenience setting, not an enforcement mechanism.

Also shipping in v2.1.243: keyless sign-in via Anthropic Console (/login), a Skipped sources line in /status for managed settings, and model + effort level visible in /tasks for subagent runs.

Sources: Claude Code changelog v2.1.243 · Model configuration docs · Enterprise managed settings guide — systemprompt.io