Making Sense of AI Models: What Powers Copilot, ChatGPT and Claude

Last updated: 19 March 2026

The Confusion Is Understandable

You have probably heard the terms thrown around: GPT, LLM, Copilot, ChatGPT, Claude. Your team might be using some of these tools already, perhaps without fully understanding what makes them different or how they relate to each other.

This post cuts through the jargon to explain what these AI systems actually are, how they connect, and what it means for your organisation.

What Is an LLM?

LLM stands for Large Language Model. Think of it as the engine that powers AI assistants. An LLM is trained on enormous amounts of text to understand and generate human language. It learns patterns, facts and ways of expressing ideas by processing billions of documents, websites and books.

When you type a question into ChatGPT or Copilot, an LLM processes your words, figures out what you are asking and generates a response. The model does not think like a human. Instead, it predicts what text should come next based on patterns it learned during training.

Different companies build different LLMs. OpenAI builds the GPT series. Anthropic builds Claude. Google builds Gemini. Each has strengths and weaknesses, which is why the choice of model matters.

The Major Players

OpenAI (GPT models)

OpenAI created ChatGPT and the GPT series of models. Their current flagship is GPT-5.4, which launched in March 2026. The GPT-5 family now includes several variants: GPT-5.4 Thinking is the most capable reasoning model, GPT-5.3 Instant is the fast everyday workhorse that replaced GPT-5.2 Instant as the default in ChatGPT, and GPT-5.2 Thinking remains available as a legacy option until June 2026. Earlier models like GPT-4o have been retired from ChatGPT entirely as of February 2026.

Anthropic (Claude models)

Anthropic builds the Claude family of models. The current generation is the Claude 4.6 family, led by Claude Opus 4.6 (released February 2026) and Claude Sonnet 4.6 (released February 2026). Claude Haiku 4.5 remains available as the fast, cost-efficient option. Claude models are known for strong performance on coding, complex reasoning and maintaining context over very long conversations, with Opus 4.6 and Sonnet 4.6 both supporting a one-million-token context window. They are built with a particular focus on being helpful while avoiding harmful outputs.

Google (Gemini models)

Google builds Gemini, which powers their AI features across Search, Workspace and other products. The current generation is Gemini 3, with Gemini 3.1 Pro now serving as Google's most advanced model. Gemini 3 Flash is the default in the Gemini app, offering fast responses at lower cost. Google also offers Gemini 3 Deep Think mode for the most complex reasoning tasks. Gemini is designed for multimodality, meaning it can work across text, images, video and audio.

Meta (Llama models)

Meta releases the Llama series as open-weight models, meaning organisations can download and run them on their own infrastructure. This appeals to businesses wanting more control over their AI deployment.

ChatGPT vs Copilot: What Is the Difference?

This is where many people get confused. ChatGPT and Copilot are both AI assistants, but they serve different purposes and work in different ways.

ChatGPT is OpenAI's consumer product. You can use it directly at chat.openai.com or through the mobile app. It is powered by GPT-5.3 Instant and GPT-5.4 Thinking (depending on your plan and query complexity) and is designed for general use: answering questions, writing content, analysing data, generating images and more. ChatGPT works with whatever information you provide in your conversation plus what it can find on the web.

Microsoft Copilot is different. While it also uses GPT models from OpenAI (through Microsoft's Azure infrastructure), Copilot is designed specifically for the workplace and integrates deeply with Microsoft 365. The key difference is what data it can access.

Microsoft Copilot: Web vs Work

Understanding the difference between Copilot's modes is essential for making sense of what it can do for your organisation.

Web mode grounds Copilot's responses in information from the internet. Ask it about industry trends, competitor analysis or general knowledge, and it searches the web to inform its answer. This is similar to how ChatGPT works when browsing is enabled.

Work mode is where Copilot becomes genuinely powerful for businesses. In Work mode, Copilot can access your organisation's data through Microsoft Graph. This includes your emails, Teams chats, SharePoint files, meeting recordings and other content stored in Microsoft 365. When you ask Copilot to summarise last week's project updates, it can pull from actual emails you received, meetings you attended and documents your team shared.

This is where a concept called RAG comes in. Retrieval Augmented Generation means the AI does not just rely on what it learned during training. Instead, it retrieves relevant information from your organisation's data and uses that to generate more accurate, contextual responses. When Copilot in Work mode answers a question, it is grounding its response in your actual business content.

Free Copilot Chat vs Paid Microsoft 365 Copilot

Microsoft offers two tiers of Copilot access, and the differences matter.

Microsoft 365 Copilot Chat (sometimes called the free tier) is included with your Microsoft 365 subscription at no extra cost. It provides:

• AI chat grounded in web data
• Enterprise data protection for your prompts and responses
• Features like Copilot Pages, file upload and image generation
• Side-by-side access when working in apps like Word and Excel

However, Copilot Chat cannot search across your organisation's Microsoft Graph data as part of the chat experience. It only knows about your work content if you upload a file or copy content into your prompt.

Microsoft 365 Copilot (the paid licence) includes everything in Copilot Chat plus:

• Chat grounded in your work data (files, meetings, emails, chats)
• Embedded Copilot features in Word, Excel, Outlook, PowerPoint and Teams
• Access to advanced reasoning agents like Researcher and Analyst
• Priority access to capabilities and newer models

The paid licence is where Copilot transforms from a capable web assistant into an AI that actually understands your business context.

Standard vs Priority Access

Even within the free Copilot Chat tier, Microsoft distinguishes between standard and priority access. Standard access means features like file upload, image generation and newer models are subject to service capacity. During busy periods, these features may be limited or unavailable.

Priority access, included with the paid Microsoft 365 Copilot licence, ensures you get consistent access to these capabilities without waiting for capacity.

Model Choice in Copilot

Here is something that surprises many people: Microsoft Copilot is not locked to a single AI model.

As of March 2026, the Copilot model picker in Microsoft 365 Copilot gives you direct control over which model handles your request. The options are:

• Auto (the default) lets Copilot decide how long to think, routing simple requests to a fast model and complex requests to a deeper reasoning model.
• Quick Response prioritises speed, answering right away using a fast model.
• Think Deeper forces Copilot to spend more time reasoning through your request for a more thorough answer.

Under the "More" menu, you can also select specific GPT models directly:

• GPT-5.3 Quick Response and GPT-5.4 Think Deeper are the latest options.
• GPT-5.2 Quick Response and GPT-5.2 Think Deeper remain available as legacy choices.

This means you can match the model to the task. A quick factual question? Use Quick Response. Drafting a complex strategy document or analysing a large dataset? Switch to Think Deeper or GPT-5.4 Think Deeper for the best results.

Beyond the model picker, Microsoft has expanded model choice even further. The Researcher agent, which helps with complex multi-step research tasks, can now be powered by either OpenAI's reasoning models or Anthropic's Claude. Users with the paid Copilot licence can choose which model to use for deep research work.

In Copilot Studio, where organisations build custom agents, Claude Sonnet 4.5 and Claude Opus 4.5 are available alongside OpenAI options, and GPT-5.2 is available for agent orchestration. This means businesses can select the best model for specific use cases, or even mix models within multi-agent workflows.

For most users, the Auto setting will handle model routing intelligently. But for organisations with specific requirements or those building custom solutions, having explicit model choice provides valuable flexibility.

A Note for Technical Teams

For organisations with advanced requirements, it is worth knowing that custom AI solutions are possible. Through Azure AI Foundry and Copilot Studio, businesses can build agents that use their preferred models, connect to their own data sources and integrate with line-of-business systems.

Some organisations even bring their own models or fine-tune existing ones with their proprietary data. This is beyond what most businesses need, but the capability exists for those with specific requirements.

What This Means for Your Organisation

The AI landscape can feel overwhelming, but here are the practical takeaways:

If you are using Microsoft 365, you already have access to Copilot Chat at no extra cost. It is a solid starting point for web-based AI assistance with enterprise data protection.

If you want AI that understands your business, the paid Microsoft 365 Copilot licence unlocks work-grounded chat and embedded features across your apps. This is where the productivity gains become significant.

If you are evaluating different AI tools, understand that ChatGPT and Copilot serve different purposes. ChatGPT is excellent for general use, Copilot is designed for work within the Microsoft ecosystem.

If you are building custom solutions, model choice now extends well beyond OpenAI. Claude and Gemini models are available for specific scenarios, and the Copilot model picker puts that choice directly in users' hands.

The AI tools are becoming more capable every month. The key is understanding what you actually have access to and using it effectively.

Our Microsoft Copilot training helps teams understand which tier they need and how to get the most from whichever licence they have. Not sure where to start? Our AI readiness assessment helps organisations figure out what makes sense before investing in licences or training. Get in touch and we can talk through what would work best for your situation.