Playbooks · Teams
Copilot's model-policy API is GA at four levels
Written November 13, 2026
GitHub's model-policy GET/PATCH endpoints are generally available at four levels. How an org decides which models stay selectable.
What changed
On October 19, GitHub announced that the REST API can now retrieve and update Copilot model policies, and that the GET and PATCH endpoints are generally available. A model policy is the setting that decides which models people in an enterprise can pick in Copilot. Until this release, changing that list meant working through settings screens at each level of the organization. Now the same decision is a resource an admin can read and write with an API call, which means it can live in a script, a review process, or a scheduled job like any other configuration.
The timing is the context. Copilot's model list has been changing in waves, with older models retiring and replacements arriving on their own schedule. Each retirement forces the same admin question: is the replacement model enabled for our people, and is the retiring one still selectable anywhere? A settings screen answers that one organization at a time. An API answers it everywhere at once.
The four levels
The changelog lists one GET and PATCH pair per level:
- Enterprise:
/enterprises/{enterprise}/copilot/model-policies - Enterprise teams:
/enterprises/{enterprise}/teams/{team_slug}/copilot/model-policies - Agent apps:
/enterprises/{enterprise}/copilot/apps/model-policies - Organization:
/orgs/{org}/copilot/model-policies
GET returns the policy in force at that level. PATCH updates it. Because the endpoints exist at every level, an admin can read the whole stack before changing any of it, which is the order that matters: a change at the enterprise level and a change at the organization level do not mean the same thing, and the delegation mode decides which one governs.
Delegation decides who governs
GitHub's changelog describes two delegation shapes. Enterprise policies can delegate model control to organizations in organization mode, or to teams and apps in enterprise teams mode. In the first shape, the enterprise sets the frame and each organization decides its own list. In the second, control sits with enterprise teams and agent apps instead. Either way, the level that holds the delegated control is the level whose policy people actually feel in the model picker.
That is the part to settle before writing any automation. A PATCH sent to a level that does not hold control in the current mode will not produce the picker your users see. Read the enterprise policy first, learn which mode is live, and only then change the level that governs.
A sensible working pattern
- Start with GET at every level you manage, and keep the responses. They are your record of what was selectable, where, and when you looked.
- Make the change at the governing level only, with the smallest PATCH that expresses it: one model on, one model off.
- GET again after the change and compare. The second read is what turns a hopeful API call into a known state.
- When a retirement wave lands, work from the replacement side: enable the successor model in policy first, then retire the old one, so nobody opens Copilot to an empty picker. The retirement-side work is covered in Re-pin Copilot models before retirements.
Limits
- The changelog announces the endpoints and the delegation modes. Questions of token scopes, rate limits, and per-endpoint fields belong to GitHub's REST API documentation, which the changelog links for getting started.
- A policy controls availability, not quality or price. Enabling a model makes it selectable; whether it should be your team's default is a separate decision.
- The four levels are not four copies of one switch. Treating them as interchangeable is the main way to change a policy and see nothing happen.
Related playbooks
- Copilot code review now has an API
- Set a spending policy before Copilot billing does
- Anthropic's model-protection rule is now in force
- Claude Code pools budgets and reports costs via API
Sources
Back to: Teams playbooks hub · Also: Re-pin Copilot models before retirements · Copilot Agent Merge: watch a PR and fix its blockers