In Microsoft Copilot, you usually cannot manually change the model; model choice appears only in Copilot Studio or Azure OpenAI, where you pick the deployment before you build the copilot. If you keep looking for a model switch in the chat UI, you will waste time on a setting that is not there and may use the wrong tool for long prompts or high-volume work. This guide shows where model choice appears, how to check what Copilot is using, and which model fits each task, cost band, and context size.
Can you change the model in Copilot?
Yes, but only in the surfaces that expose model selection. In Copilot Studio, the prompt builder has a Model control at the top, and that is where you switch among supported generative AI models. In Microsoft 365 Copilot, Copilot Chat, and Edge Copilot, users do not get a manual model picker.
Where model selection is available
The place to change models is the Copilot Studio prompt builder. The same prompt builder is used for prompts in Copilot Studio, Power Apps, and Power Automate, and licensing or credits depend on the product. If you are building prompts, that is where you compare model families before you publish.
Where model selection is not available
For everyday chat surfaces, Copilot chooses the model automatically. The UI may show a single chat box, not a model menu. If you need manual control for a prompt, move the work into Copilot Studio or the related builder surface instead of searching inside the chat pane.
How auto model selection works
Microsoft says Copilot can choose the model and processing level automatically based on the task. That helps when you want speed and less setup, but it also means the system decides the tradeoff for you. If the result is slower than expected or costs more in credits, that usually points to task complexity, not a hidden user toggle. In auto-selection mode, there is no user-facing way to verify the exact model for a specific chat turn inside the ordinary Copilot UI.
Copilot model matrix
This matrix shows where choice exists, what model options appear, and how the main models compare on token window, cost tier, and best use. It is the fastest way to match a prompt to the right model without guessing.
| Copilot surface | Available model choices | Default model | Token window | Credit tier | Best-use case |
|---|---|---|---|---|---|
| Copilot Studio prompt builder | GPT-4.1 mini, GPT-4.1, GPT-5 chat, GPT-5 reasoning, Claude Sonnet 4.6, Claude Opus 4.6 | GPT-4.1 mini | Varies by model; context window is enforced per call | Basic, standard, or premium; prompts consume Copilot Credits | Prompt design, testing, and deployment |
| Power Apps prompt builder | Generative AI models exposed in the shared prompt builder | GPT-4.1 mini | Varies by model; context window is enforced per call | Depends on product licensing or credits | App prompts that need controlled model choice |
| Power Automate prompt builder | Generative AI models exposed in the shared prompt builder | GPT-4.1 mini | Varies by model; context window is enforced per call | Depends on product licensing or credits | Workflow prompts with predictable cost control |
| Microsoft 365 Copilot, Copilot Chat, Edge Copilot | No normal manual model picker for most users | Auto-selected by Copilot | Managed by the service | Included in the service plan or tenant setup | General chat and assisted work |
Model-by-model detail: GPT-4.1 uses the standard rate and supports input up to 128K tokens. GPT-5 chat uses the standard rate, is trained on data up to around September 2024, and supports up to 128K tokens. GPT-5 reasoning supports context up to 400K tokens and uses the premium rate. Claude Sonnet 4.6 is an external model from Anthropic, uses the standard rate, and supports context up to 200K tokens. Claude Opus 4.6 is external from Anthropic, uses the premium rate, and supports context up to 200K tokens. (Microsoft Learn)
What you'll need
- Access to Copilot Studio, Power Apps, or Power Automate if you want manual model choice.
- A tenant with the right licensing or credit setup for the product you are using.
- A prompt that is long enough to justify model selection, such as a multi-part analysis task.
- Permission to use the relevant model tier in your environment.
- A clear target for the prompt: speed, cost control, or deeper context handling.
How to change the model in Copilot Studio
Open the prompt builder, then use the Model selector at the top of the builder. That is the switch point. In Copilot Studio, the same prompt builder is used for prompts in Copilot Studio, Power Apps, and Power Automate, and prompts consume Copilot Credits.
- Open the prompt builder. Go to the prompt-building surface in Copilot Studio, Power Apps, or Power Automate. The model selector appears at the top of the prompt builder, not in the chat response area.
- Choose the Model menu. Select Model at the top of the builder and pick the family that fits the task. If you do not see it, the tenant may not expose that model or you may be in the wrong product surface.
- Match the prompt to the tier. Use GPT-4.1 or GPT-5 chat for standard work; GPT-5 reasoning for very long or complex prompts; Claude Sonnet 4.6 or Claude Opus 4.6 when the Anthropic option is available and allowed.
- Check token fit. Keep the prompt and any included history within the model’s context window.
- Save and retest. Send a short validation prompt after switching models so you can compare response shape, speed, and credit use before you build a larger flow.
Say your prompt includes a long policy draft and several examples. GPT-5 reasoning gives you much more room at 400K tokens when the combined input is large and the answer needs more structured reasoning.
Suppose you only need a short, routine summary for a flow in Power Automate. GPT-4.1 is a better starting point when speed and lower credit use matter more than deep context.
Why is the model dropdown missing or locked?
If the dropdown is missing or locked, the usual causes are product-surface limits, tenant policy, or licensing and credit constraints. The fix is rarely inside the prompt itself. First verify that you are in Copilot Studio, Power Apps, or Power Automate, because the model picker is not a general chat control. No Copilot chat surface, including Microsoft 365 Copilot, offers manual model selection.
License or credit limits
Copilot Studio prompts consume Copilot Credits, and those credits are used for prompt runs in that builder. If the environment has no usable credits, or the product licensing does not permit that model tier, the selector may be disabled or simplified.
- Check the tenant setup. Confirm that the product and environment are licensed for prompt use and that credits are available for the model tier you want.
- Switch to a lower tier. Try GPT-4.1 first, then move up to GPT-5 chat only if the task needs more capability. That reduces the chance of a locked premium option.
Region or tenant availability
Model availability varies by region. If a model is listed in documentation but not in your tenant, the region may not expose it yet, or the environment may not be enabled for that family.
- Confirm the environment region. Check the tenant or environment region in the admin settings for the product.
- Retry after a policy refresh. If the region is correct, re-open the builder and reload the page so the picker can refresh its available models.
Product surface mismatch
Many users look in Microsoft 365 Copilot, Copilot Chat, or Edge Copilot and expect a model picker. Those surfaces do not expose one. Move the prompt into Copilot Studio or the shared prompt builder instead.
- Open the correct builder. Use Copilot Studio, Power Apps, or Power Automate, then locate the Model control at the top of the prompt builder.
- Do not search the chat UI. If the chat surface has no model selector, that is expected for most users.
What affects which Copilot model is used?
Three things drive model choice: the context window, the credit tier, and the task type. The context window is the limit for each model call, enforced by the model provider, and total turn usage can exceed a single call window when the conversation spans multiple turns. That means long prompts need a model with enough room.
Context window limits
Short prompts usually do fine on GPT-4.1. Large policy drafts, long histories, or multi-file prompts need more headroom.
Credit tier and pricing
Basic, standard, and premium rates matter because they affect how far your credits go. GPT-4.1 and GPT-5 chat are standard, and GPT-5 reasoning and Claude Opus 4.6 are premium. If you are running many prompt calls, the cheaper tier often makes more sense.
Task type and response tradeoffs
Use smaller models for fast extraction, simple rewriting, and short summaries. Use larger or premium models when the prompt needs broader context, more careful reasoning, or better handling of complex instructions. Copilot’s automatic mode is designed to balance accuracy, speed, and cost when you do not need manual control.
Common first-timer mistake: choosing a premium model for a simple task and spending credits faster than needed. A better approach is to start with GPT-4.1, then move up only when the answer quality or context limit is no longer enough.
Common mistakes
- Looking for model choice in Microsoft 365 Copilot, Copilot Chat, or Edge Copilot instead of the prompt builder.
- Assuming Claude is a normal built-in choice in every Copilot surface. In this setup, Claude appears in Copilot Studio prompt builder, not as a general consumer chat toggle.
- Ignoring the token window and pasting more context than the model can accept in one call.
- Using a premium model by default when GPT-4.1 is enough.
- Forgetting that Copilot Credits and licensing affect what is available in the builder.
Frequently asked questions
How do I change the model in Microsoft Copilot?
Use the prompt builder, not the chat box. In Copilot Studio, the Model selector sits at the top of the builder and lets you choose among supported models before you run the prompt. If the selector is missing, check tenant access, region availability, and credit status.
Can I change the default model in Microsoft Copilot?
You can choose a different model only where the builder exposes that option. The default model is GPT-4.1, but changing it depends on the product, licensing, and available model list. In general chat surfaces, you do not set a personal default model.
What models are available in Microsoft Copilot Studio?
Copilot Studio prompt builder can offer GPT-4.1, GPT-5 chat, GPT-5 reasoning, Claude Sonnet 4.6, and Claude Opus 4.6. The exact list depends on region and tenant setup. GPT-4.1 is the default model, and prompts consume Copilot Credits.
Why can't I select a different model in Copilot?
Most often, you are in a surface that does not support manual selection. The next most common causes are missing credits, licensing limits, or regional availability. Go to the prompt builder in Copilot Studio, Power Apps, or Power Automate and check the Model control at the top.
Does Microsoft Copilot automatically choose the model?
Yes, when you are using Copilot’s automatic mode. Microsoft says Copilot can choose the model and processing level automatically based on task needs. That setup balances accuracy, speed, and cost, but it does not give you manual selection inside the ordinary chat UI.
Is Claude available in microsoft Copilot?
Claude is available in Copilot Studio prompt builder as an external Anthropic model family, not as a normal built-in choice in every Copilot app. Claude Sonnet 4.6 uses the standard rate and supports 200K tokens, while Claude Opus 4.6 uses the premium rate and also supports 200K tokens.