A model is a large language model (LLM) — a statistical system trained on vast amounts of text that predicts the next token given what came before. After training, it becomes very good at one thing: completing sequences of tokens — words, punctuation, code — which, scaled up massively, produces outputs that look like reasoning and code generation. A model has no memory between sessions; each new conversation starts completely fresh unless the product layer above injects context into the prompt. You can swap models in GitHub Copilot because the model is just one component of the stack — change the engine, and the product around it stays the same.

The key thing to understand

⚠️ The model is not GitHub Copilot. Copilot is a product that calls models.

When you switch to “Claude Sonnet 4.6” in Copilot, you are changing which model Copilot sends your prompt to. You are still using GitHub Copilot as your assistant. The model is the engine under the hood; Copilot is the car.

Model examples

These are some of the models available in GitHub Copilot (as of early 2026):

ModelProviderNotes
GPT-5.4OpenAIStrong general-purpose reasoning
Claude Sonnet 4.6AnthropicWell-suited for long, complex tasks
Gemini 2.5 ProGoogleStrong at long context windows
Grok Code Fast 1xAISpeed-optimised for coding tasks

Model availability changes frequently — check the GitHub Copilot model docs for the current list.

What a model can’t do

  • No memory — it doesn’t retain anything between sessions
  • No tools — it cannot browse the web, run code, or read your files on its own
  • No autonomous action — it responds to prompts; it doesn’t initiate anything
  • No product context — it knows nothing about your repo, IDE, or billing plan unless told

Everything a model appears to “know” about your situation was injected into the prompt by the product layer above it — the AI coding assistant.

Dive deeper

To change your model in GitHub Copilot, open Copilot Chat in VS Code and use the model picker at the bottom of the chat panel. On GitHub.com, the picker appears at the top of the conversation. For in-depth guidance on which model to choose for which task — and how model selection differs across chat, agent mode, and the coding agent — read Models in GitHub Copilot.

Understanding what a model can’t do is exactly what makes agents make sense. An agent uses a model as its reasoning core but adds the ability to take action. Read What is an Agent? to see how.

Related Concepts

📍 Layer 0 — The Model · Back to the AICA Onion

What’s next?

Updated: