The AI Models Powering Microsoft 365 Copilot

Last updated: 19 March 2026

There's More Than One Model Behind Copilot Now

If you've opened Microsoft 365 Copilot recently and noticed a dropdown in the top-right corner of Copilot Chat, you're not imagining things. Microsoft has been quietly rolling out model choice, and most users now have access to multiple AI models and response modes.

The problem is that none of this is explained particularly well inside the product. So if you've been wondering what the difference is between the default experience and GPT-5.4, what Auto, Quick Response and Think Deeper actually do, where GPT-5.3 fits in, or why there's now a Claude option appearing, this post breaks it all down in plain English.

GPT-5: The Default Model Powering Copilot

As of late 2025, GPT-5 is the standard model that powers Microsoft 365 Copilot for all licensed users. It became the default in a phased rollout between October and November 2025, and by now it's mandatory across web, desktop and mobile. You can't opt out of it, and you don't need to do anything to enable it.

GPT-5 introduced something Microsoft calls dynamic model routing. Behind the scenes, Copilot looks at your prompt and automatically decides which variant of the model to use. Simple questions get routed to a fast, high-throughput model that responds quickly. More complex prompts get sent to a deeper reasoning model that takes more time but produces more thoughtful, structured answers.

For most everyday use, GPT-5 with its automatic routing does a solid job. If you're summarising emails, drafting messages, asking quick questions or getting meeting recaps, the default experience handles it well without you needing to think about models at all.

The Response Modes: Auto, Quick Response and Think Deeper

Alongside the model selector, Microsoft has introduced a response mode selector that gives you direct control over how much reasoning Copilot applies before answering. You'll find this near the prompt area in Copilot Chat, and there are three options:

• Auto: This is the default mode. Copilot assesses your prompt and decides for itself whether to give you a fast answer or take more time to reason through it. For most users, this is the sensible choice to leave selected. It balances speed and quality seamlessly.
• Quick Response: This tells Copilot to prioritise speed. It answers right away without extended reasoning. Use it when you're drafting emails, polishing text, translating content, getting quick definitions or iterating fast on short-form work.
• Think Deeper: This is the opposite. Copilot takes additional time to plan its response, reason through the problem and check its own work before answering. Use it when you're analysing complex documents, building business cases, working through multi-step problems, comparing options or generating detailed reports where accuracy matters.

Your mode selection persists across chats, so once you've set it, it stays until you change it.

The Optional Upgrades: GPT-5.2, GPT-5.3 and GPT-5.4

This is where it gets interesting. In addition to the default GPT-5 experience, Microsoft has made OpenAI's newer models available as optional upgrades you can select from the model dropdown in the top-right corner of Copilot Chat.

GPT-5.2

Rolling out since December 2025, GPT-5.2 delivers better instruction following, improved maths and coding performance and stronger multi-step reasoning. It also handles significantly longer context, meaning it's better at maintaining coherence across long chat threads or large documents.

When paired with the Think Deeper mode, GPT-5.2 is a powerhouse. On OpenAI's GDPval benchmark (which measures AI performance across real-world professional tasks), GPT-5.2 Thinking achieved a 70.9% win or draw rate against human experts, compared to GPT-5's 38.8%. If you require genuine reasoning rather than pattern matching, selecting GPT-5.2 with Think Deeper is worth the extra wait.

GPT-5.2 now appears in the model picker as both GPT-5.2 Quick Response and GPT-5.2 Think Deeper, and is considered a legacy option now that newer models are available.

GPT-5.3

On 3 March 2026, Microsoft began rolling out GPT-5.3 Instant. It appears in the model picker as GPT-5.3 Quick Response.

GPT-5.3 replaced GPT-5.2 Instant as the everyday fast model. It delivers more accurate responses, stronger writing and better web synthesis, meaning it does a better job of combining search results with its own knowledge rather than just dumping links at you.

Real-world testing shows its quality is highly dependent on the task:

• Where it shines: GPT-5.3 is exceptional at structuring chaotic information. If you feed it a messy brain dump and ask for a prioritised board meeting agenda, it makes incredibly smart decisions about what to include, what to defer and how to structure the output.
• Where it falls short: For natural, warm human communication (like turning meeting notes into a client email), GPT-5.3 can feel a bit stiff and clinical. Surprisingly, leaving Copilot on the default Auto mode often produces a much better, ready-to-send email.

GPT-5.4: The New Flagship

As of mid-March 2026, Microsoft has started rolling out GPT-5.4 Thinking to Microsoft 365 Copilot users. It appears in the model picker as GPT-5.4 Think Deeper and is now the most capable reasoning model available in Copilot.

GPT-5.4 is OpenAI's latest frontier model. It combines the coding capabilities of GPT-5.3-Codex with improved reasoning, agentic workflows and a one-million-token context window. In practical terms, this means it can hold significantly more information in its working memory when reasoning through complex tasks.

Where GPT-5.3 focuses on speed and everyday quality, GPT-5.4 focuses on depth. Microsoft describes it as better at working through complex, multi-step tasks with more clarity and consistency, producing stronger first drafts without the usual back-and-forth. It is also more token-efficient than GPT-5.2 when reasoning, meaning it solves problems using fewer internal tokens while delivering better results.

Use GPT-5.4 Think Deeper when you need Copilot to go deep: analysing lengthy contracts, building detailed financial models, comparing strategic options across multiple documents or working through technical problems that require sustained reasoning. For quick everyday tasks, the default Auto mode or GPT-5.3 Quick Response will still be faster and more practical.

Why Not Just Use the Newest Model All the Time?

If GPT-5.4 is the most capable model available, why wouldn't you just leave it selected permanently? There are two main reasons:

1. Newer doesn't mean better for every task. The default GPT-5 Auto routing often writes warmer, more natural-sounding emails in a fraction of the time. GPT-5.3 Quick Response is excellent for structured information tasks. GPT-5.4 Think Deeper is overkill for a quick email draft and will take noticeably longer to respond. Match the model to the complexity of the task.
2. Compute resources and fair use limits. Deeper reasoning models consume significantly more tokens internally as they reason, plan and self-check. Microsoft 365 Copilot operates on a fair use basis. While enterprise users have effectively unlimited usage, Microsoft applies rate limiting behind the scenes to ensure fair allocation. If your whole team spent the day hammering GPT-5.4 Think Deeper for every simple message, you'd likely hit throttling faster than if everyone stuck to Auto.

Save the manual model selection for the work that genuinely benefits from it: strategy documents, complex analysis, detailed reports and multi-step problem solving.

Claude Is Now Available Too, But There's a Catch for UK Organisations

Microsoft has also started offering Anthropic's Claude models (Claude Sonnet 4.5 and Claude Opus 4.5) within Copilot Studio and the Researcher agent.

As of January 2026, Anthropic has been onboarded as a Microsoft subprocessor, meaning Claude usage is covered under Microsoft's Product Terms and Data Protection Addendum (DPA). Admins can manage this from the Microsoft 365 admin centre under Copilot > Settings > Data access > AI providers operating as Microsoft subprocessors.

The UK and EU Problem

Here's the important detail for UK and EU organisations: Anthropic's models are currently excluded from the EU Data Boundary and in-country processing commitments. All Claude processing happens on Anthropic's infrastructure in the United States. Your data leaves Microsoft's managed environment, gets processed in the US and then returns. Because of this, Microsoft has toggled Claude to Off by default for organisations in the EU, EFTA and the UK.

If your organisation has strict GDPR or data residency requirements, you need to think carefully before enabling Claude. The DPA covers the data protection, but the physical processing location introduces international data transfer considerations that your compliance team must assess. You can opt-in, but it requires an active decision by an admin.

A Quick Summary

Here's the simple version of how it all fits together:

• GPT-5 is the default model. It powers Copilot automatically and routes your prompts to the right variant behind the scenes.
• Auto lets Copilot decide how much reasoning to apply. Leave it here for 80% of your daily work, especially natural writing tasks.
• Quick Response prioritises speed. Use it for fast drafting, translations and simple questions.
• Think Deeper prioritises quality. Use it for analysis, strategy and complex reasoning tasks.
• GPT-5.3 Quick Response is the latest fast model, great for structuring information and everyday analytical tasks.
• GPT-5.4 Think Deeper is the most capable reasoning model now available in Copilot, best for complex, multi-step work that demands depth and accuracy.
• GPT-5.2 remains available as a legacy option in both Quick Response and Think Deeper modes.
• Claude is available via the admin centre but is Off by default for UK and EU tenants due to US-based data processing.

The key takeaway is that you now have real choices about how Copilot works for you. Model access varies depending on whether you are on Copilot Basic or Premium, so it is worth knowing which tier you are on. Do not assume "newer" means "better" for every prompt. The default Auto experience is fantastic, but knowing exactly when to reach for GPT-5.4 Think Deeper or GPT-5.3 Quick Response can make a meaningful difference to your output quality.

For accounting firms and law firms in particular, understanding which model handles which task type is a practical skill that directly affects your bottom line.

If you'd like help getting your team trained on these new features through our Microsoft Copilot training, or need guidance on the admin settings and compliance implications, book a consultation with us.