Using GPT-5 on Microsoft

GPT5 Integration in Microsoft Experiences

Key Concepts:

• GPT5: Unified OpenAI model with real-time routing between high-throughput and deep reasoning models.
• Model Unification: Combining specialized capabilities into a single model.
• Real-time Routing: Dynamically selecting the optimal model (GPT5, GPT Mini, GPT Nano, GPT Chat, or even GPT4 variants) based on the prompt.
• Deep Reasoning: Utilizing a model trained with reinforcement learning to think before answering, involving planning and work checking.
• Reasoning Effort: Configurable parameter (minimal, low, medium, high) controlling the amount of internal reasoning, impacting token usage and response time.
• Verbosity: Configurable parameter (low, medium, high) controlling the amount of output and output tokens generated.
• PhD Reasoning Level: High level of knowledge and capability.
• Multimodal Input: Accepting both text and images as input.
• Context Window: Large input (up to 272,000 tokens) and output (up to 128,000 tokens) capacity.
• Azure AI Foundry: Platform for accessing and using base models, including GPT5 variants.
• GitHub Copilot: AI-powered coding assistant with edit, ask, and agent capabilities, now enhanced with GPT5.
• Copilot Studio: Low-code platform for building agents and copilots, leveraging GPT5 for agent orchestration and prompting.
• Microsoft Copilot: Interaction experience with a "smart mode" powered by GPT5 for intelligent task handling.

1. Introduction to GPT5

GPT5 is not just the latest version of the OpenAI GPT model; it unifies previously separate capabilities and reasoning abilities distributed across different models. This model unification allows for real-time routing, where the system chooses between a high-throughput model for faster responses and a deeper reasoning model for complex tasks.

2. Deep Reasoning Capabilities

GPT5's deep reasoning is enhanced through reinforcement learning, encouraging the model to "think before answering." Prompts like "walk through the problem step by step," "evaluate multiple options," or "check your work" trigger the reasoning option.

• API Configuration: When using the API, developers can configure the "reasoning effort" (minimal, low, medium (default), high), which affects token usage and response time. Higher reasoning effort increases lag before the first response token.
• Verbosity Control: The API also allows setting "verbosity" (low, medium (default), high) to control the amount of output and commentary generated, especially useful in coding scenarios.

3. Reasoning Level and Focus Areas

GPT5 boasts a "PhD reasoning level," making it a highly knowledgeable and capable partner. Development has focused on improving performance in writing, coding, and health-related tasks.

• Examples:
  • Writing: Generating a complete app from a single, well-structured prompt.
  • Health: Understanding medical reports and providing guidance (not a replacement for medical professionals).
  • Agentic Tasks: Excelling in RFP analysis, compliance audits, and financial modeling.

4. Multimodal Input and Context Window

GPT5 accepts both text and images as input (multimodal input) but does not generate images. It supports very long context windows: up to 272,000 tokens for input and 128,000 tokens for output (shared between internal reasoning and the final output).

5. Safety and Improvements

GPT5 includes improved safety measures, better jailbreak protection, and reduced hallucinations.

6. GPT5 in Azure AI Foundry

Azure AI Foundry provides access to base models, including GPT5, GPT Mini, GPT Nano, and GPT Chat.

• GPT5: Standard model for complex reasoning, broad knowledge, heavy coding, and multi-step agentic tasks.
• GPT5 Mini: Cost-optimized reasoning and chat, balancing speed, cost, and capabilities.
• GPT5 Nano: For very high throughput tasks and simple instruction following/classification.
• GPT5 Chat: For multi-turn, multimodal interactions with context maintenance.

The model router in Foundry automatically selects the most optimal model based on the prompt, considering cost optimization.

• Azure AI Foundry Integrations: Content safety capabilities, virtual network integration, private endpoints, Entra ID authentication, managed identities, and fine-tuning options.

7. GPT5 in GitHub Copilot

GitHub Copilot integrates GPT5 across its edit, ask, and agent capabilities, improving code quality, maintainability, and reasoning for complex tasks. Users can select GPT5 within Copilot.

8. GPT5 in Copilot Studio

Copilot Studio, a low-code platform, leverages GPT5 for agent orchestration and prompting. Users can select GPT5 Auto (switches between GPT5 Chat and GPT5 Reasoning) or pure GPT5 Reasoning. This functionality is also available in Power Apps.

9. GPT5 in Microsoft Copilot

Microsoft Copilot features a "smart mode" that uses GPT5 to think deeply or quickly based on the task. It automatically routes queries to the appropriate model (chat-based or advanced reasoning) based on complexity. This option is also available in Copilot Chat and M365 Copilot.

10. Conclusion

GPT5 is rapidly being integrated into various Microsoft experiences, offering significant improvements in reasoning, coding, and task execution. Its availability across Azure AI Foundry, GitHub Copilot, Copilot Studio, and Microsoft Copilot provides users with diverse opportunities to leverage its capabilities. The real-time routing and configurable parameters allow for optimized performance and cost management.