Choosing and pinning models in GitHub Copilot

Published 2026-05-22 by Owner

On May 17, GPT-5.3-Codex became the base model for Copilot Business and Enterprise. “Base” means the model you get when you do not choose one — so if you have never opened the model picker, your default just changed. Here is how to know which model is writing your code and how to take the choice back where you can.

Base model vs. your choice

For an individual on Pro, Pro+, or Free, the base-model change does not apply — you keep selecting models yourself. For anyone on Business or Enterprise, the org’s base model is the default, and an org admin sets it. Establish which side of that line you are on first, because it decides whether “pin a model” is something you do or something your admin does. If you are on an enterprise plan and disagree with the base, the conversation is with whoever administers your org, not with a local setting. That is not a bug — centralized model policy is the point of an enterprise plan — but it does mean the lever you reach for depends entirely on your seat type, and reaching for the wrong one wastes an afternoon. Confirm your plan in your GitHub account’s billing settings before you go looking for a control that your tier may not expose.

See which model is active

In VS Code, the active chat model shows in the Copilot Chat model picker — that is the authoritative place to read it, not the command line. The official CLI extension is still worth installing so you can drive Copilot from the terminal; confirm it is present with:

gh extension install github/gh-copilot
gh copilot --version

That reports the extension version, not the model. For the model itself, read the Chat picker: the model it names is the one answering your prompts, and the inline-completion model is configured separately and can differ, so check both rather than assuming one covers your whole session.

Pick a model per session

In Copilot Chat, the model picker selects the model for the current conversation only; it does not change the org base. Selecting Gemini 3.5 Flash (GA in Copilot since May 19) or any other available model is per-session. The Copilot settings in VS Code live under the real github.copilot namespace:

{
  "github.copilot.enable": { "*": true }
}

A per-default model-pinning setting may exist in your Copilot version; confirm the exact key in VS Code Settings under “Copilot” before relying on it, rather than trusting a key copied from an article.

What the multipliers mean

GPT-5.3-Codex carries a 1× premium-request multiplier; GPT-4.1 ran at 0× until it deprecates with usage-based billing on June 1. A multiplier is what each request costs against your premium-request allowance: a 0× model is effectively free against that meter, a 1× model spends one unit per request, and a higher-multiplier model spends more. The shift from a 0× default to a 1× default means the same usage that cost nothing against the meter last month starts drawing it down this month. Once flex billing starts, the base model you are routed to has a direct cost, so “which model is the default” becomes a budget question and not only a quality one. The cost mechanics are in managing Copilot costs.

Detect silent changes

The May 17 switch happened without most developers touching anything, and the next one will too. To avoid being surprised, make the model visible in your own workflow: note the active model in the description of any AI-assisted PR, and check the picker after each Copilot update. The broader decision of which model to standardize on is in the AI model selection decision; the argument for why provenance matters is you didn’t pick this model, and the change that started it is GitHub Copilot makes GPT-5.3-Codex its enterprise base model.