Which AI Model Powers GitHub Copilot? (2026)
GitHub Copilot isn't powered by a single model. It's powered by a rotating shelf of the best frontier models on the market — and you can pick.
Last updated:
GitHub Copilot is powered by a rotating selection of frontier language models, primarily from OpenAI (GPT-4o, GPT-4.1, the o-series) and increasingly from Anthropic (Claude Sonnet) and Google (Gemini). For inline completions, GitHub picks the default model. In Copilot Chat, you choose the active model in the toolbar. Copilot Enterprise customers get priority access to the highest-tier model variants.
If you're trying to figure out which model is actually running when Copilot suggests code, the answer is: it depends on which surface you're using, which plan you're on, and which model you've picked in Chat. Here's the current breakdown for 2026.
Inline Code Completions
Inline completions — the ghost text that appears as you type — are generated by a GitHub-tuned variant of a frontier model from OpenAI. GitHub does not currently let you change the inline-completion model in the UI; they manage it server-side to balance latency, quality, and cost. The active model has rolled forward from Codex → GPT-3.5-class → GPT-4-class → GPT-4o-class, and continues to upgrade silently.
Copilot Chat — User-Selectable Models
In Copilot Chat (VS Code, Visual Studio, JetBrains, GitHub.com), there's an explicit model picker. The currently exposed shelf includes OpenAI GPT-4o, GPT-4.1, the o-series reasoning models, Anthropic Claude Sonnet, and Google Gemini variants. The exact list changes month to month; check the picker in your IDE for the current set.
What Copilot Enterprise Adds
Copilot Enterprise customers get priority access to the highest-tier variants of these models, higher rate limits, and longer context windows. For organizations relying on long-context refactors or deep PR review, the Enterprise tier's model access alone often justifies the $39/seat/month.
Why GitHub Uses Multiple Vendors
A single-model strategy is risky in a fast-moving market. By exposing OpenAI, Anthropic, and Google models, GitHub lets engineers pick the right tool for the task — Claude tends to win on long-context refactors, GPT-4-class models win on inline completion latency and reasoning, Gemini wins on certain multimodal tasks. The model picker future-proofs Copilot against any single lab pulling ahead.
How to See Which Model Is Active
In VS Code Copilot Chat, the active model is displayed in the bottom of the chat input. Click it to switch. In Visual Studio it's in the Copilot Chat toolbar. In GitHub.com Copilot Chat, look at the model selector in the chat header. If you don't see a picker, you're likely on a Free or Pro tier with a fixed default.
Does Your Code Train These Models?
On Copilot Business and Copilot Enterprise, no — your prompts and suggestions are not used to train the underlying foundation models, and code snippets are not retained. Copilot Enterprise can index your private repos to power knowledge-base chat, but those indexes are scoped to your organization and removed when the subscription ends.
If you want Copilot Enterprise's priority model access for your team, see the Enterprise plan or contact us for a volume quote.
Frequently Asked Questions
Common questions related to this guide — sourced from real searcher queries.
GitHub Copilot is powered by a rotating selection of frontier language models, primarily from OpenAI (GPT-4o, GPT-4.1, o-series) and increasingly from Anthropic (Claude Sonnet) and Google (Gemini). Copilot Enterprise users get priority access to the highest-tier model variants. The active model is shown — and switchable — in the Copilot Chat toolbar.
GitHub Copilot uses multiple LLMs: by default a GPT-class model from OpenAI for inline completions, plus user-selectable models in Copilot Chat including GPT-4o, GPT-4.1, o-series, Claude Sonnet, and Gemini. Enterprise plans get priority access to the highest-tier variants.
Yes — GitHub Copilot exposes GPT-4-class models from OpenAI (currently including GPT-4o and GPT-4.1) in Copilot Chat. Default inline completions also run on GPT-4-class infrastructure. The exact build is managed server-side by GitHub and upgrades silently.
Yes — GitHub Copilot Chat exposes Claude Sonnet as a selectable model alongside OpenAI's GPT models and Google Gemini. Pick the active model in the chat toolbar. Copilot Enterprise customers get priority access to the highest-tier Claude variants.
Yes. Copilot Free has request-rate limits and exposes a smaller model picker. Copilot Pro ($10/month) removes the rate caps and unlocks higher-tier models in Chat. For teams, Copilot Business ($19/seat/month) adds policy controls, public-code-match blocking, and IP indemnity on top of the same model access.