Copilot Enterprise Can Now Default to Any Model — Including Kimi 3 — and Teams Can Override It

GitHub announced on 2026-09-02 something that matters to anyone buying AI tooling for a company: enterprise administrators can now set any model as the default for new GitHub Copilot conversations, and different teams can carry different defaults.

1. The problem this solves

Once a company has bought Copilot, the hard part is not who may use it — it is which model people actually end up using. The model picker offers a long list, everyone grabs whatever, and the result is:

• A bill spread across a dozen models that nobody can explain
• Cost per model varying wildly, chosen purely by individual habit
• No way to roll out the model you just finished evaluating, other than asking people nicely to switch

2. What changed

Per the announcement:

1. Enterprise admins can designate a default model for new conversations via managed settings. The examples given include Kimi 3, GPT-6 Astra and Gemini 3.8 Flash — “any default model” is literal, not marketing.
2. Per-team granularity: set the model key as overridable in team-mappings.json and enterprise teams can choose their own preferred defaults, while everyone else inherits the enterprise-wide setting.
3. Surfaces covered: the GitHub Copilot app, Copilot CLI, and Visual Studio Code.
4. Plans: generally available with Copilot Business and Copilot Enterprise.

3. What it means per role

For platform and IT admins: this is a governance lever, not a lockdown. Set a sensible cost/quality default company-wide, then mark model as overridable so the data team can move to a stronger reasoning model while the front-end team runs something fast and cheap. Because it lives in a config file, the change is reviewable in a pull request rather than clicked into a console.

For developers: your default changes, but you can generally still switch models inside a conversation. What changes is which model you get when you do not choose — which, given how few people change defaults, is the most effective intervention available.

For finance and procurement: this is the first step to treating model mix as a managed variable. Copilot cost conversations used to stop at seat count; now they can cover which model is the default and who may reach for the expensive one.

4. Our take: the default is the policy

On the surface this is a settings toggle. Underneath it reflects a maturing stage of enterprise AI buying: from “should we buy it” to “how do we manage usage and model mix”.

Two caveats worth stating:

• Default is not exclusive. The announcement covers the default for new conversations, not a lock. If your goal is to forbid a specific model, that belongs to model access controls, not this setting.
• That config file carries real power. team-mappings.json decides what hundreds of engineers use every day. Changes to it need review and history, or an unexplained bill spike becomes untraceable.

If your company is currently asking whether Copilot spend is justified, put this near the top of the list: fix the default model first, so the bill has a baseline to compare against.

Source: GitHub Changelog, 2026-09-02, verified 2026-09-07. Follow GitHub’s documentation for configuration details. Copilot plans and allowances are tracked on our GitHub Copilot tool page.