Compare AI Copilots — Models Behind the Tools
How AI copilots use large language models
An AI copilot is a product built around one or more large language models (LLMs). The model reads prompts and creates text or code. The product may add web search, file access, memory, or coding tools.
That is why asking what LLM a tool uses can have more than one answer. A service may route different tasks to different models. It may also change its model lineup by plan, region, or user choice.
Names can be confusing, too. Some describe a model family, while others name a whole product system. Microsoft’s Prometheus, for example, links GPT models with search and product context.
- Copilot uses Microsoft’s Prometheus system with OpenAI GPT models.
- Claude and Gemini are model families and assistant products.
- Cursor and Perplexity can offer several model choices.
- Model access can change as products and plans evolve.
Check the product’s model picker or help pages when a precise version matters. A tool’s brand name alone may not reveal which model handled a task.
What LLM does Microsoft Copilot use?
Microsoft Copilot is based on Prometheus, a system that uses OpenAI GPT large language models. Prometheus connects those models with Microsoft’s search tools and context. It is not a standalone base LLM.
Microsoft described Prometheus as a way to ground responses in Bing search results. Search can help a reply draw on current web pages. The model still creates the response, while the surrounding system supplies useful context.
So, what LLM does Copilot use? The short answer is OpenAI GPT models within Microsoft’s Prometheus system. Which LLM does Copilot use can depend on the Copilot product and task. Consumer Copilot, Microsoft 365 Copilot, and developer tools may not use one fixed model in every case.
Work features may bring in files and app data that the user can access. That does not mean the model has learned those files. The service can pass relevant information for a request, then return a response. Microsoft’s description of its AI-powered Bing explains the early Prometheus approach.
Look at the product and feature, not just the Copilot name. That gives a more useful answer.
Compare models in similar AI tools
ChatGPT is OpenAI’s assistant and uses models from the GPT family. The first public ChatGPT release used GPT-3.5. That answers what LLM does ChatGPT use for its early version, but not for every current plan or feature.
Claude uses Anthropic’s Claude models. Anthropic puts a strong focus on safety and alignment, which means shaping model behaviour to follow intended goals. Google’s Gemini models power the Gemini assistant. Gemini replaced the Bard name, so older references to Bard may describe an earlier product stage.
Code and search tools can give users more model choice. Cursor lets users select models from providers such as OpenAI, Anthropic, and Google. Perplexity combines language models with web search, and its Sonar family is built for search-based answers.
| Tool | Model or system | Key point |
|---|---|---|
| Microsoft Copilot | Prometheus with OpenAI GPT models | Search and product context shape responses |
| ChatGPT | OpenAI GPT family | GPT-3.5 powered the first public release |
| Claude | Anthropic Claude family | Anthropic focuses on safety and alignment |
| Gemini | Google Gemini family | Gemini replaced the Bard name |
| Cursor | Several provider models | Choices depend on plan and feature |
| Perplexity | Sonar and other model options | Web search is central to its answers |
| Lovable | Models used in its app-building service | Available models may change |

What models do coding and research tools use?
What LLM does Cursor use? Cursor supports models from several providers, rather than relying on one public, fixed model. Its own product layer connects those models with code context and editing features. Which LLM does Cursor use can depend on the model selected and the task.
What LLM does Replit use? Replit’s AI features use models through its coding tools, and the available options can vary. What LLM does Claude Code use? Claude Code uses Anthropic’s Claude models to help users work with code. In both cases, features and model access can shift over time.
What LLM does Perplexity use? Perplexity uses its Sonar models and other model options for search-led answers. Which LLM does Perplexity use may depend on the selected mode or plan. Search and source retrieval are part of the product, not simply a model name.
What LLM does Lovable use? Lovable uses models within a service that turns natural-language requests into app changes. Its model setup may change, so users should check current product details. The product’s app-building tools also shape the result.
What LLM does Augment Code use? Augment Code offers AI coding help, but its model choices and routing can change. What LLM does Manus use? Manus is an AI agent product, and its underlying model setup may vary by feature or service update. Avoid assuming either product has one permanent model.

Claude, Gemini, NotebookLM, and Character AI
What LLM does Claude use? Claude is powered by Anthropic’s Claude model family. Different Claude versions can offer different levels of speed and reasoning. The product name alone does not identify the exact version used for each request.
What LLM does Gemini use? Google’s Gemini models power Gemini, Google’s assistant. Gemini also replaced Bard as the product name. Model access can depend on the feature, account, or plan.
What LLM does NotebookLM use? NotebookLM is a Google research tool that uses Google’s Gemini models to work with supplied sources. Its source-based design matters as much as the model. It can answer from the material a user adds, rather than acting only as a general chatbot.
What LLM does Character AI use? Character AI relies on its own model systems for interactive character chats. The company does not present the service as a simple front end for one publicly named model. Its model details and features may change.
What LLM does GitHub Copilot use? GitHub Copilot offers models from more than one provider in some features and plans. It is part of the broader Copilot family, but it is a distinct coding product. Check its current model picker for the options available to your account.

How to check a tool’s current model
Model details can go stale quickly. Product names may stay the same while providers add models, change defaults, or route requests differently. Treat any single model name as a snapshot unless the vendor confirms it still applies.
Start with the product’s model picker, help centre, or release notes. Then check whether the page describes a default model, an optional model, or the system behind the whole product. These are different claims.
- Confirm the exact product and feature you use.
- Check the model picker for your plan or account.
- Read recent vendor notes for model changes.
- Separate the base model from search, tools, and app context.
This approach also helps compare tools fairly. One may use a strong model but add little context, while another may pair a model with search or code access. The model is only one part of the result.
AI copilots will keep adding model choices and routing features. Some will use several models behind one interface. For users, the practical questions are which model is available, what data reaches it, and how well the full product handles the task.