NemoClaw Knowledge Wiki

Using GPT5 with Microsoft copilot

3 min read

https://www.youtube.com/watch?v=XA78I5V-BC4 The video announces the integration of GPT-5 into Microsoft 365 Copilot and Copilot Studio, highlighting its key advancements and practical implications. Key Features of GPT-5:

  • Smarter, Faster, and Built-in Thinking: GPT-5 is described as OpenAI’s best AI system to date, offering significant leaps in intelligence, performance across various tasks (coding, math, writing, health, visual perception), and deeper reasoning capabilities.
  • Unified System with Real-Time Routing: Unlike previous models that required users to choose the appropriate model, GPT-5 acts as a “real-time router.” It intelligently understands the user’s prompt and automatically selects the best model for the task: Smart, High-Throughput Model: For common or routine questions, it prioritizes speed for quick, succinct responses. Deeper Reasoning Model: For complex or open-ended questions, it engages “chain of thought reasoning” to plan, gather relevant context, and check its work for more comprehensive and accurate answers.

Demonstrations in Microsoft 365 Copilot:

  1. Simple Query Example (Addressing Unfair Manager Treatment): Without GPT-5: Copilot provided general steps based on workplace experts and policies (Anti-Discrimination, Health and Safety). It offered reasonable suggestions but lacked specificity. With GPT-5 Enabled: Copilot responded much faster and more directly. It immediately focused on the relevant “Grievance Policy” and provided internal steps according to company policy, including informal and formal procedures, confidential support, and legal rights. The speaker noted the higher quality and speed for basic questions.

  2. Complex Query Example (Evaluating a Grievance Policy for Clarity): Without GPT-5: Copilot provided a breakdown of the policy’s strengths (clear purpose, emphasis on informal resolution, supportive principles) and general recommendations for improving clarity (structure, language simplification, procedural steps, consistency, accessibility). It was “not bad” but somewhat generic. With GPT-5 Enabled: Copilot engaged in “chain of thought reasoning,” explicitly detailing its process (searching for grievance policy, preparing to evaluate clarity, fetching content, checking tool availability, evaluating snippets, planning clarity improvements). The resulting recommendations were significantly more specific and practical. It cross-referenced other relevant policies (Whistleblowing, Anti-Discrimination, Disciplinary, EEO) and offered actionable suggestions like making “what counts as a grievance” explicit, presenting the process as a time-bound journey, clarifying roles, specifying submission channels, strengthening confidentiality, and tightening cross-policy navigation. The speaker emphasized this was “much more specific” and “way better.”

Availability for Copilot Studio Users:

  • Users building custom agents in Copilot Studio can also access GPT-5 by choosing it as their agent’s model in the settings, specifically the “GPT-5 Reasoning (Experimental)” option.

Conclusion: The speaker concludes that GPT-5 brings significant improvements in response quality and speed, especially for complex queries requiring deeper reasoning. Users are encouraged to try the “Try GPT-5” button in Copilot to experience the difference firsthand.

Graph View