This document provides an overview of the different models available in Copilot and guidance on when to use each one effectively. Copilot is free for everyone, offering a variety of models for different tasks.
- Not all models are available in all modes. In "ask" mode, you have access to more models than in "agent" mode.
- A reasoning model excels at tasks like creating plans, project requirements documents, and outlining project steps.
- Good for high-level planning, which can then be passed to an agent model for implementation.
- General-purpose agent model with unlimited requests in Copilot.
- Better at tool calling compared to its predecessor (40).
- Suitable for agentic workflows but may require explicit instructions to continue or avoid making plans instead of executing tasks.
- The Claude family (3.5, 3.7, and 4.0) are excellent for agent mode.
- Key difference between the versions is the training date, with 4.0 being the most recent.
- Using Claude 40 provides prom caching, allowing for more requests before reaching limits compared to 3.5 or 3.7.
- A multimodal model that understands text, images, and sound.
- Works well in agent mode and is useful in many workflows.
- A smaller reasoning model.
- May be slower in agent mode.
- Not as effective as 41 or the Claude models.
| Model | Type | Description | Best Use Cases |
|---|---|---|---|
| o3 | Reasoning | Excels at creating plans and project requirements documents. | High-level planning for projects, outlining steps. |
| GPT 4.1 | Agent | General-purpose agent model with unlimited requests. | Agentic workflows, tool calling. |
| Claude (3.5/3.7/4.0) | Agent | Fantastic performance in agent mode; Claude 4.0 has the most recent training data and supports prom caching for more requests. | Complex agent tasks requiring reliability and efficiency. |
| Gemini 2.5 Pro | Multimodal Agent | Understands text, images, and sound; offers versatile capabilities. | Workflows that benefit from multimodal input (text and images). |
| o4 Mini | Reasoning | A smaller reasoning model that may be slower in agent mode. | Basic reasoning tasks, but consider other models for better performance. |
- Consider starting with Claude 4,0 for most agent-related tasks due to its updated training and prom caching benefits.
- Use Gemini 2,5 Pro if your workflow benefits from multimodal inputs such as images and text.
- For high-level planning and project outlining, leverage the reasoning capabilities of the o3 model before passing tasks to agent models.