GitHub Copilot models: what's available, plan access and how to switch

As of October 2026, GitHub Copilot supports 27 AI models from six providers: OpenAI (GPT-5.3-Codex through GPT-6.1 Sol), Anthropic (Claude Haiku, Sonnet, Opus and Fable), Google (Gemini 3.8 Flash and Gemini 4 Argon), Microsoft (MAI-Code-1.1-Flash), Moonshot AI (Kimi K3) and xAI (Grok 4.5 to 4.7). Copilot Free and Copilot Student get these models through auto model selection only; paid plans can also pick a model in the model picker.

Everything below comes from GitHub's supported models reference, read on 26 October 2026. GitHub notes that "Model availability is subject to change," so check that page before you rely on a specific model.

Which AI models does GitHub Copilot support?

GitHub's table lists every model below as generally available (GA). Claude Opus 4.8 (fast mode) still carries a "preview" label in its name.

A few footnotes matter in practice:

• GPT-5.4 nano is "currently only available in the Codex Visual Studio Code extension (Copilot Pro+ only) and is not available in Copilot Chat." GitHub's per-plan table marks it as included on Pro+ and Max.
• Claude Sonnet 4.6 was retired on 1 September 2026 but stays available to individual subscribers on annual Pro and Pro+ plans. Monthly subscribers don't get it.
• Claude Fable 5 and 5.1: GitHub says Anthropic retains prompts and outputs by default when these models are used, to operate safety classifiers. Customers can request zero data retention through the end of 2026 under a time-bound exemption.
• GPT-5.3-Codex is the long-term support (LTS) model for Copilot Business and Enterprise, designated on 18 March 2026 and supported for one year. Auto falls back to it "in the event no other models are available."

Which models does each Copilot plan include?

Copilot Free and Copilot Student don't get a model picker. GitHub's note is direct: "Copilot Student and Copilot Free users have access to models through auto model selection only."

On paid plans, the split looks like this, based on the per-plan table on the plans page:

So the jump from Pro to Pro+ is mostly about the most capable models: the Claude Opus and Fable line, the GPT Sol and Astra models, and Gemini 4 Argon. Max has the same models as Pro+, except Claude Sonnet 4.6, with a larger AI credit allowance.

In Business and Enterprise, admins decide what members can use. A default-availability policy controls whether new GA models switch on automatically, but some models are always off until an admin enables them: pre-GA models, open weight models such as Kimi K3, and Claude Fable 5 and 5.1, which GitHub says are "not covered by GitHub's data retention agreement." In Copilot Business, the organization must also grant members the ability to switch models in Copilot Chat.

How does auto model selection work?

Pick Auto in the model picker and Copilot routes each task to a model based on task complexity and real-time system health and availability, per GitHub's auto model selection page. The supported models reference lists the Auto pool as GPT-5.6 Luna, Sol and Terra, GPT-6 Astra, Luna and Sol, GPT-6.1 Sol, Claude Haiku 4.5, Claude Opus 4.8, 5 and 5.5, Claude Sonnet 5 and 5.5, Gemini 3.8 Flash and MAI-Code-1.1-Flash, limited by your plan and policies.

In VS Code, Copilot CLI and the GitHub Copilot app, Auto has three tiers:

Auto has a cost benefit too. On a paid plan you get a 10% discount on model costs when you use Auto in Copilot Chat, Copilot CLI, the Copilot app or Copilot cloud agent. GitHub has also run larger, time-limited Auto discounts for Copilot Max, so check the auto model selection page for any current offer.

On individual plans, Auto may serve evaluation models, which can appear under codenames. GitHub warns they "may perform worse than other models on security-related, or other categories of prompts." You can turn them off in your GitHub settings under Copilot AI controls, with the Evaluation models in Copilot auto model selection setting.

To see which model answered, hover over the response in Copilot Chat; Copilot CLI prints it in the terminal.

How do you change the model in Copilot?

GitHub's model switching guide covers each client:

• VS Code: open Copilot Chat and pick a model from the dropdown at the bottom of the chat view. VS Code's default keyboard shortcuts open the language model picker with ⌥⌘. on macOS or Ctrl+Alt+. on Windows and Linux. Manage Models in the same dropdown adds models from other providers, which may ask for an API key.
• Visual Studio: View > GitHub Copilot Chat, then the model dropdown at the bottom right. Multi-model chat needs Visual Studio 2022 version 17.12 or later.
• JetBrains IDEs: open Copilot Chat from the status bar icon and pick a model. For reasoning models, hover over the model and choose a level in the Thinking Effort submenu.
• Copilot CLI: type /model in a session. /model --global saves a default for future sessions. You can also start with a flag or environment variable.
• Copilot cloud agent: choose a model when you assign an issue to Copilot, mention @copilot in a pull request comment, or start a task from the agents tab or panel, GitHub Mobile or Raycast. Elsewhere, Auto is used.

# start Copilot CLI with a specific model
copilot --model=claude-opus-5.5

# or let Copilot choose
COPILOT_MODEL=auto copilot

GitHub's CLI reference lists claude-sonnet-5.5 as the default model. See the Copilot CLI guide for the rest of its commands.

Changing the chat model does not change the model behind inline suggestions. In VS Code, run GitHub Copilot: Change Completions Model from the Command Palette for that.

Which models work in which client?

From GitHub's per-client table:

• GitHub.com doesn't offer Gemini 3.8 Flash, Gemini 4 Argon, the Grok models or Claude Opus 4.8 (fast mode).
• Eclipse, Xcode and JetBrains IDEs don't offer GPT-5.6 Luna, Sol or Terra, or Claude Sonnet 4.6.
• Copilot CLI, VS Code and Visual Studio offer the widest range.
• The 1 million token context window is available "in Visual Studio Code and Copilot CLI only," on supported models such as Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Sol and Kimi K3.
• Configurable reasoning levels work in VS Code, Copilot CLI and Copilot cloud agent.

If a model is missing from your picker, update your IDE and the Copilot extension first: some models need a minimum version.

How do models affect your AI credits?

Since 1 June 2026, Copilot bills on GitHub AI Credits, not premium requests. Each interaction uses input, output and cached tokens priced by model, and the total converts at 1 AI credit = $0.01 USD. Some sample rates per 1 million tokens from GitHub's models and pricing page, as of October 2026:

The spread is large: Claude Fable 5 output costs 100 times more per token than GPT-6 Luna. Longer context and higher reasoning levels also burn more credits, so GitHub recommends the regular context window and reasoning level by default.

Code completions and next edit suggestions are not billed in AI credits and stay unlimited on paid plans. Pro and Pro+ subscribers on existing annual plans that still use request-based billing have separate, legacy model multipliers. For plan allowances, see the GitHub Copilot pricing guide.

Which Copilot model should you use?

GitHub doesn't name a single best model. A practical approach:

1. Start with Auto. It picks a model per request and costs 10% less on paid plans.
2. Use a lightweight model for quick questions. GitHub's pricing tables mark GPT-6 Luna, GPT-5.6 Luna, Claude Haiku 5.5 and MAI-Code-1.1-Flash as "Lightweight."
3. Switch to a "Powerful" model for hard problems, such as large refactors or tricky bugs in agent mode: Claude Opus 5.5, GPT-6 Sol or GPT-6.1 Sol, on Pro+ or higher.
4. Watch your usage dashboard after a few long agent sessions.

GitHub's AI model comparison gives task-based suggestions if you want a second opinion. To compare with other tools, see Claude Code models.

Keep prompts portable across models

Models come and go: GitHub retired GPT-5.4, GPT-5.5 and Gemini 3.7 Flash on 19 October 2026 alone. Prompts that work well are worth keeping outside any one tool. Kipt stores your prompts and Agent Skills as Markdown in a private GitHub repo you own, with {{name}} blanks you fill in each time, so you can reuse the same prompt whichever model you pick.