Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
This article explains how to change the model version and settings in the prompt builder. The model version and settings can affect the performance and behavior of the generative AI model.
Model selection
You can change the model by selecting Model at the top of the prompt builder. The dropdown menu allows you to select from the generative AI models that generate answers to your custom prompt.
Important
In November 2025, we migrated the o3 model to the GPT-5 reasoning model. Prompts that ran on the o3 model were automatically transitioned to the GPT-5 reasoning model without action required from you. It's possible to revert temporarily to the o3 model by requesting it through a support request on prompts. This exception lasts until December 17, 2025, after which the o3 model will be permanently retired.
Using prompts in Power Apps or Power Automate consumes prompt builder credits, while using prompts in Microsoft Copilot Studio consumes Copilot Credits. Learn more in Licensing and prompt builder credits.
Available models
| GPT model | Licensing | Functionalities | Category |
|---|---|---|---|
| GPT-4.1 mini (Default model) |
Basic rate | Trained on data up to June 2024. Context allowed up to 128K tokens. | Mini |
| GPT-4.1 | Standard rate | Trained on data up to June 2024. Context allowed up to 128K tokens. | General |
| GPT-5 chat | Standard rate | Trained on data up to September 2024. Context allowed up to 400K tokens. | General |
| GPT-5 reasoning | Premium rate | Trained on data up to October 2024. Context allowed up to 400K tokens. | Deep |
| Claude Sonnet 4.5 (experimental) | Standard rate | External model from Anthropic. Context allowed up to 200K tokens. | General |
| Claude Opus 4.1 (experimental) | Premium rate | External model from Anthropic. Context allowed up to 200K tokens. | Deep |
GPT-4o mini and GPT-4o continue to be used in U.S. government regions. These models follow licensing rules and offer functionalities comparable to GPT-4.1 mini and GPT-4.1, respectively.
Learn more about external Anthropic models in Choose an external model as the primary AI model.
Licensing
In agents, flows, or apps, models used by prompts consume Copilot Credits, regardless of their release stage. Learn more in Copilot Credit management.
If you have AI Builder credits, they're consumed in priority when prompts are used in Power Apps and Power Automate, but they aren't consumed when prompts are used in Copilot Studio. Learn more in AI Builder: Overview of licensing.
Release stage
Models have different stages of release. You can try out new, cutting-edge experimental and preview models, or choose a reliable, thoroughly tested generally available model.
| Tag | Description |
|---|---|
| Experimental | Used for experimentation, and not recommended for production use. Subject to preview terms, and can have limitations on availability and quality. |
| Preview | Eventually becomes a generally available model, but currently isn't recommended for production use. Subject to preview terms, and can have limitations on availability and quality. |
| No tag | Generally available. You can use this model for scaled and production use. In most cases, generally available models have no limitations on availability and quality, but some might still have some limitations, like regional availability. Important: Anthropic Claude models are at the experimental stage, even though they don't display a tag. |
| Default | The default model for all agents, and usually the best performing generally available model. The default model is periodically upgraded as new, more capable models become generally available. Agents also use the default model as a fallback if a selected model is turned off or unavailable. |
Experimental and preview models might show variability in performance, response quality, latency, or message consumption, and might time out or be unavailable. They're subject to preview terms.
Categorization
The following table describes the different model categories.
| Mini | General | Deep | |
|---|---|---|---|
| Performance | Good for most tasks | Superior for complex tasks | Trained for reasoning tasks |
| Speed | Faster processing | Might be slower due to complexity | Slower, as it reasons before responding |
| Use cases | Summarization, information tasks, image and document processing | Image and document processing, advanced content creation tasks | Data analysis and reasoning tasks, image and document processing |
When you need a cost-effective solution for moderately complex tasks, have limited computational resources, or require faster processing, choose Mini models. It's ideal for projects with budget constraints and applications like customer support or efficient code analysis.
When you're dealing with highly complex, multimodal tasks that require superior performance and detailed analysis, choose General models. It's the better choice for large-scale projects where accuracy and advanced capabilities are crucial. Another scenario where it's a better choice is when you have the budget and computational resources to support it. General models are also preferable for long-term projects that might grow in complexity over time.
For projects requiring advanced reasoning capabilities, Deep models excel. It's suitable for scenarios that demand sophisticated problem-solving and critical thinking. Deep models excel in environments where nuanced reasoning, complex decision-making, and detailed analysis are important.
Choose among the models based on region availability, functionalities, use cases, and costs. Learn more in Feature availability by regions for prompts, and Pricing comparison table.
Model updates
| Model | Status | Retirement date | Replacement |
|---|---|---|---|
| GPT-4.1 mini | Generally available | No date yet | n/a |
| GPT-4.1 | Generally available | No date yet | n/a |
| GPT-5 chat | Generally available | No date yet | n/a |
| GPT-5 reasoning | Generally available | No date yet | n/a |
| GPT-5.1 chat | Pending availability | No date yet | n/a |
| GPT-5.1 reasoning | Pending availability | No date yet | n/a |
| Claude Sonnet 4.5 | Experimental | No date yet | n/a |
| Claude Opus 4.1 | Experimental | December 2025 | Claude Opus 4.5 |
| Claude Opus 4.5 | Pending availability | No date yet | n/a |
| o3 | Retired | December 4, 2025 | GPT-5 reasoning |
| GPT-4o mini | Retired | July 2025 | GPT-4.1 mini |
| GPT-4o | Retired | July 2025 | GPT-4.1 |
| o1 | Retired | July 2025 | o3 |
Model settings
You can access the settings panel by selecting ... > Settings at the top of the prompt builder. You can change the following settings:
- Temperature: Lower temperatures lead to predictable results. Higher temperatures allow more diverse or creative responses.
- Record retrieval: Number of records retrieved for your knowledge sources.
- Include links in the response: When selected, the response includes link citations for the retrieved records.
Temperature
The slider allows you to select the temperature of the generative AI model. It varies between 0 and 1. This value guides the generative AI model about how much creativity (1) vs deterministic answer (0) it should provide.
Temperature is a parameter that controls the randomness of the output generated by the AI model. A lower temperature results in more predictable and conservative outputs. To compare, a higher temperature allows for more creativity and diversity in the responses. It’s a way to fine-tune the balance between randomness and determinism in the model’s output.
By default, the temperature is 0, as in previously created prompts.
| Temperature | Functionality | Use in |
|---|---|---|
| 0 | More predictable and conservative outputs. Responses are more consistent. |
Prompts that require high accuracy and less variability. |
| 1 | More creativity and diversity in the responses. More varied and sometimes more innovative responses. |
Prompts that create new out-of-the-box content. |
Adjusting the temperature can influence the model’s output, but it doesn't guarantee a specific result. The AI's responses are inherently probabilistic and can vary with the same temperature setting.
Note
The temperature setting isn't available for the GPT-5 reasoning model. For this reason, the slider is disabled when you select the GPT-5 reasoning model.