Is an LLM Worth It? A Simple Way to Decide
Learn when an LLM earns its cost and how to test one for your work.
Copilot uses different AI models across products, tasks, and response modes.
Copilot does not rely on one large language model, or LLM. The model depends on which Copilot product you use, the task, and the options available in your account. GitHub Copilot offers a choice of models for coding, while Microsoft Copilot products can use different models and response modes.
That means there is no single answer to “what LLM does Copilot use?” For GitHub Copilot, check the model picker in your chat or coding tool. For Microsoft Copilot, look for the model or mode shown in the product. Access can vary by plan, region, rollout, and organization settings.
The product name matters. It is the quickest way to get the right answer.

GitHub Copilot can use models from more than one provider. The exact set changes over time. Depending on your plan and where you use Copilot, the model picker may offer models from OpenAI, Anthropic, or Google. GitHub documents the current choices in its guide to choosing AI models.
So, which LLM does GitHub Copilot use? The best answer is the model selected for that session, not a fixed product-wide model. One model may suit code completion, while another may work better for planning, explaining code, or finding a bug. Model names and access can differ between Copilot Chat, the editor, and other GitHub tools.
For example, a developer might ask one model to explain a failing test, then try another model to draft a fix. Treat model choice as a task-level decision. A model that gives a useful answer for a short code question may not be the best choice for a broad refactor.
Check the picker before comparing results. It tells you which model answered.

Consumer Copilot is the assistant for everyday questions, writing, and other general tasks. Its underlying model can change as Microsoft updates the service. Users may also see modes such as Quick response and Think Deeper, where available. These modes shape how Copilot handles a request, rather than naming a model in every case.
Quick response is suited to a direct question, a short summary, or a simple rewrite. Think Deeper is meant for tasks that benefit from more reasoning, such as weighing options or working through a multi-step problem. A deeper answer may take longer. It is not automatically better for a simple request.
Some Copilot experiences may use newer models, including GPT-5 family models. Names such as GPT-5.5 should not be read as a promise that every user has that model. Model access and labels can change by date, product, plan, and rollout. Check the model selector or product notes for the current details.
Choose a mode based on the work, not the label alone.

Microsoft 365 Copilot works within apps such as Word, Excel, PowerPoint, Outlook, and Teams. It can use language models to draft, summarize, and reason over a request. The app may also draw on work data that the user is allowed to access. That work context is separate from the model itself.
Microsoft may offer model choices or new model experiences in some Microsoft 365 Copilot features. Which options appear depends on the account, admin settings, and rollout. A feature that uses one model today may change later. For the current product details, check Microsoft's Microsoft 365 Copilot documentation.
This distinction helps when a response feels different across apps. The prompt, linked work context, app features, and model can all affect the result. If an answer seems off, confirm the source material and ask a narrower question. Do not assume that Word and Teams use identical settings.
Model choice is only one part of the answer.

Some work-focused Copilot products, including preview or early-access tools, may expose more model choices than the consumer app. An organization might see options tied to different model families or task types. Availability depends on the specific product, license, region, and access granted by an administrator.
Names such as GPT-5.6 and GPT-6 require special care. Do not assume they are generally available Copilot models just because they appear in a plan, forecast, or informal discussion. A model name is confirmed for your use only when Microsoft lists it for that product or your account displays it as an option. The same check applies to Copilot Cowork and Frontier experiences.
For a team, record the product name, model label, and date when testing an output. This makes comparisons useful even when the available choices change. It also helps separate a model change from a prompt change or a new data source.
Early access can mean more choice, but also more change.
Copilot performance depends on more than the model name. Products can use model routing, which directs a request to a suitable model or processing path. A short factual question may need a fast answer. A code task or a complex plan may benefit from deeper reasoning.
Routing can happen behind the scenes, so users may not see every choice. The system may also use tools, search, or permitted work data to answer. These features can affect speed, detail, and accuracy. They do not make every answer correct.
When you need a more reliable result, name the task and share useful context. For code, include the error, relevant file, and expected behavior. For a work summary, point Copilot to the right source. Then check claims that could affect a decision.
In short, which LLM Copilot uses depends on the Copilot product and the choice available for your task. Use the in-product model picker as your source of truth. It is more reliable than a broad claim about one model powering all Copilot experiences.
Learn when an LLM earns its cost and how to test one for your work.
Compare leading LLMs by task, cost, speed, and how well they fit your tools.
See how LLMs work, where they fit in AI, and what they can do.