At‑a‑glance summary
GPT‑5 is OpenAI’s newest generation model and it’s rolling out in Microsoft 365 Copilot. It’s built to choose the right way to answer your question automatically, aiming for better accuracy and fewer wrong turns. For small businesses, that means clearer outcomes with less faff—while keeping the same enterprise‑grade security posture you already rely on.
First things first: what’s a “model”?
An AI model is a pattern‑spotter that reads and writes information to help with everyday work—drafting, summarising, planning. GPT‑5 adds an auto‑chooser that picks the right ‘thinking mode’ for your task, so you don’t have to.
GPT‑5 in simple terms
GPT‑5 is a unified system with a built‑in router that automatically chooses between a quicker, high‑throughput model and a deeper reasoning model, based on the complexity of your prompt and what you’ve asked it to do.
Why this is better: it removes the burden from users to know, guess or test which model is best for a given query—the system makes that call in real time, improving consistency and saving time. It also aims for fewer hallucinations, stronger instruction‑following and less sycophancy (over‑agreeable behaviour).
Why OpenAI dialled down the “sycophancy”
Earlier behaviour sometimes skewed towards flattering agreement. GPT‑5 is tuned to challenge when needed—asking clarifying or tougher questions when you’re planning or strategising—so it behaves more like a thoughtful colleague than a cheerleader. OpenAI’s own post‑mortem explains why this matters and how they recalibrated for usefulness over surface‑level warmth.
The GPT‑4o uproar: what happened — and why it’s a sign to lean in, not sit out
When GPT‑5 launched, OpenAI initially removed direct access to GPT‑4o. Users pushed back—hard—so OpenAI reinstated 4o for subscribers. Beyond the headlines, the reaction showed how deeply these tools have become part of people’s daily habits. For SMEs, the takeaway isn’t to avoid AI; it’s to adopt it deliberately: build literacy, add light guardrails, and measure outcomes.
For Microsoft 365 users, these changes don’t affect continuity—Copilot remains governed by your organisation’s admin controls and security policies, so feature shifts at OpenAI won’t disrupt your workflows.
GPT‑5 in Microsoft 365 Copilot (what you should know)
- Availability: GPT‑5 is rolling out in Microsoft 365 Copilot and Copilot Studio.
- How to enable it: In Copilot Chat, use “Try GPT‑5”. It applies for that conversation; if you start a new chat, you’ll need to turn it on again, keeping you in control.
Security & reliability: steady posture, smarter outputs
- Security & privacy: Inside Microsoft 365, Copilot works with your content and keeps it within your organisation’s boundaries. Your prompts and responses aren’t used to train the foundation models. GPT‑5 also aims to reduce hallucinations, especially when deeper reasoning is needed.
- Reliability: OpenAI reports GPT‑5 is significantly less likely to hallucinate than earlier models, especially in its deeper “thinking” mode.
OpenAI GPT‑5 vs Microsoft 365 Copilot: how they relate
OpenAI builds the models; Microsoft 365 Copilot applies them to your work inside your Microsoft 365 tenant with internal data context, admin controls and compliance. In short: you get the latest model capability, anchored in your organisation’s data, kept within your 365 boundaries, and governed by your policies. That’s the pragmatic route for SMEs who want value without sacrificing control.
Adoption signals & market context
We’re sharing these numbers to show pace. They help answer, “Is this a fringe trend, or something our business should plan around?”
- ChatGPT usage surged: Sam Altman indicated ChatGPT’s weekly active users roughly doubled from ~400m to ~800m within just a few months between February and August 2025 (TED remarks reported by Forbes and VentureBeat). This signals mainstream usage, not an early‑adopter bubble. (ChatGPT Statistics & Total Users (2025): DAU & MAU Data)
- Google’s token usage doubled in two months: Google reported processing ~480 trillion monthly tokens in F 2025 (I/O) and then over ~980 trillion by July 2025—about a 104% increase. Tokens are the chunks of data models read and write; more tokens ≈ more data processed/generated, a useful proxy for real‑world AI use. (Q2 earnings call: CEO’s remarks)
Get started
- In Copilot Chat, run the same prompt twice—once with GPT‑5 turned on (use Try GPT‑5) and once without—then compare the answers for clarity, accuracy and usefulness. Repeat with a slightly more complex version of the prompt to see when the deeper reasoning pays off.
- Unsure where to start? Try some of our featured prompts (insert link)
Learn fast with our Copilot training days
Want practical, hands‑on experience and prompt patterns that actually work? Join our next Microsoft 365 Copilot Training Day. You’ll leave with a better understanding of the capabilities of Copilot, hands on experience with real world scenarios and the confidence to implement Copilot across your business.