Choosing the right model helps you move faster, spend less, and get better results. Cursor supports all top-tier models. Most models can complete any task, but they behave differently, and those differences matter.

​
How models differ

Models are trained in different ways and respond in unique styles. Some “think before they code,” while others jump straight into writing. Some take initiative and move quickly, while others take time to understand your instructions before acting. Here are some dimensions to consider:
  • Assertiveness: Some models (like gemini-2.5-pro or claude-4-sonnet) are confident and make decisions with minimal prompting.
  • Curiosity: Others (like o3 or claude-4-opus) take time to plan or ask questions to understand context more deeply.
  • Context window: Some models can process more of your codebase at once, which is useful for large-scale tasks.

​
Why it matters

Each model has different strengths. Some excel at rapid implementation, while others are better suited to planning and exploring options. Choosing the right model allows you to:
  • Get faster output
  • Receive higher quality suggestions
  • Optimize your usage and costs
Just like working with humans, each model interprets prompts differently. You’ll develop intuition over time: how each one reads, thinks, and acts. This helps you know which model fits your task best.

​
Model behavior

One way to think about model behavior is by how much initiative it takes.

​
Thinking models

These models infer your intent, plan ahead, and often make decisions without needing step-by-step guidance.
  • Ideal for when you want the model to run with the task
  • Less prompting required, though sometimes more opinionated
  • Can make bigger changes than you might expect
Examples:
  • claude-4-opus
  • gemini-2.5-pro
  • o3 (designed for complex reasoning)
Use these when you’re exploring ideas, refactoring broadly, or want the model to act more independently.
Thinking models are usually more expensive than non-thinking models. Review model pricing.

​
Non-thinking models

These models wait for explicit instructions. They don’t infer or guess and are ideal when you want to steer the output directly.
  • Ideal for precise, controlled changes
  • Require more prompting, but behave more predictably
  • Easier to guide, revise, and fine-tune
Examples:
  • claude-4-sonnet
  • gpt-4.1
Use these when you want tight control, need consistent behavior, or are working on well-defined tasks.

​
Choosing by style

Many users pick their preferred model based on interaction style more than task type. Some like assertive models that take the lead. Others prefer ones that wait for instruction. claude-4-sonnet, gemini-2.5-pro, and gpt-4.1 can all serve as reliable daily drivers - it comes down to how much control you want.

​
How to select

Cursor gives you access to a curated set of high-performing models. You can choose based on multiple factors, here are some common ones:

​
1. Prompting style

If you prefer to…Models
Be in control, give clear instructionsclaude-4-sonnet, gpt-4.1
Let the model take initiativeclaude-4-opus, gemini-2.5-pro, o3

​
2. Task type

Use this table to match your task to a suitable model:
| Task | Models | | --- | --- | | Directed changes | claude-4-sonnet, gemini-2.5-pro | | Codebase navigation/search | gemini-2.5-pro, claude-4-opus, o3 | | Planning or problem-solving | claude-4-opus, gemini-2.5-pro | | Complex bugs or deep reasoning | o3 |
o3 is designed for complex, ambiguous problems. It is powerful but also slower and more resource-intensive, which makes it better suited for occasional use.

​
Selection tree

These are subjective recommendations. You should choose the model that works best for you.

​
Auto-select

Auto keeps you in flow by picking a reliable model from the pool above (excluding o3). It does not route based on task type, but is a solid default if you are unsure which to choose.

​
Save what works

Once you find combinations that work well, like specific prompts paired with certain models, you can save them as Custom Modes. These let you:
  • Preselect a model
  • Add custom instructions
  • Reuse the setup for future tasks

​
Updates

This section tracks major revisions and updates to this guide.
DateChanges
Late May 2025Updated recommendations for newer models. Simplified categories as capabilities improve.
Early May 2025Initial version covering model selection guidance, behavior patterns, and selection criteria

​
Takeaways

  • You should choose the model that works best for you
  • Some models take initiative. They are useful for exploration, planning, and tasks where you want the model to contribute ideas.
  • Others follow instructions closely. They are useful for precision, predictability, and tasks where you want direct control.
  • claude-4-sonnet, gemini-2.5-pro, and gpt-4.1 are all strong daily drivers. Your choice depends on interaction style.
  • o3 is designed for the hardest problems.
  • Auto-select is a safe default if you are unsure.
  • Save effective setups as Custom Modes to streamline your workflow.