Choosing and Switching AI Models
Choosing and Switching AI Models
Cursor is not locked to a single AI model. It supports multiple frontier models and lets you switch between them at any time — in chat, in inline editing, and in agent mode. Knowing which model to reach for, and when, is one of the higher-leverage skills you can develop as a Cursor user.
In this lesson, you will learn which models are available in Cursor, how to switch between them, what each model is best suited for, and how to think about cost and usage limits when choosing a model.
What You'll Learn
- Which AI models are available in Cursor and how to access them
- How to switch models in the chat panel and Cmd+K inline edit
- The tradeoffs between speed, quality, and cost across models
- Which tasks benefit from powerful models vs faster, lighter ones
- Cursor's monthly request limits and how different plans affect model access
- How to bring your own API keys to use models outside Cursor's quota
- A practical model selection strategy for different coding scenarios
Models Available in Cursor
Cursor gives you access to a range of frontier AI models. The exact lineup changes as new models are released, but the core categories have been consistent.
Claude Models (Anthropic)
Anthropic's Claude models have become a favorite among developers for their strong code understanding, nuanced instruction-following, and long context windows.
Claude Sonnet is Cursor's primary workhorse for code. It balances quality and speed well — it generates thoughtful, accurate code without the longer wait times of the most powerful models. For most coding tasks, Claude Sonnet is the right default choice.
Claude Opus is Anthropic's most capable model. It reasons more carefully, produces more nuanced explanations, and handles complex multi-step problems better. The tradeoff is speed and cost — it is slower and uses more of your monthly request quota.
GPT-4o (OpenAI)
GPT-4o is OpenAI's flagship multimodal model. It handles code competently, particularly for tasks where vision input is useful (like analyzing a screenshot of UI or a diagram). Some developers prefer GPT-4o for certain tasks because of its familiarity or slightly different reasoning style.
o1 and o3 Models (OpenAI)
OpenAI's reasoning models (o1, o3, and their mini variants) use an extended thinking process before generating output. They are significantly better at complex algorithmic problems, mathematical reasoning, and tasks where correctness under constraint matters more than speed. They are considerably slower and more expensive than standard models.
Gemini Models (Google)
Google's Gemini models are also available in Cursor. Gemini 1.5 Pro has an extremely large context window, making it useful for tasks that involve reading many files at once. Gemini Flash is a fast, lightweight option when you need quick completions.
Cursor's Own Models
Cursor also ships cursor-fast and cursor-small, which are smaller, faster models optimized for Tab completion and other low-latency tasks. You generally do not select these manually — Cursor uses them automatically for instant completions where speed is critical.
How to Switch Models
Switching models in Cursor takes a few seconds and the interface is consistent across all AI features.
In the Chat Panel (Cmd+L)
The chat panel shows the currently selected model in a dropdown at the top of the input area. Click it to see the full model list and select a different one. Your selection persists for that chat session.
Chat panel → Model dropdown (top of input) → Select model
You can also switch mid-conversation. If you started with Claude Sonnet and reach a problem that needs deeper reasoning, switch to Claude Opus on the next message. The conversation history carries over.
In Inline Edit (Cmd+K)
When you trigger Cmd+K, a floating input bar appears with a model selector. The same model dropdown is available here. Inline edits often use a faster model by default, but you can switch to a more capable model if the task is complex.
Cmd+K → Model dropdown (inline bar) → Select model
In Agent Mode (Cmd+I)
Agent mode (also called Composer) has its own model selector. Since agent tasks can be long-running, your model choice here has a larger impact on both quality and cost. Claude Sonnet is typically the best default for agent tasks.
Setting a Default Model
In Cursor Settings, you can set a default model for each AI feature. Most developers settle on one model as their primary default and switch only when a specific task demands it.
Understanding the Tradeoffs
Choosing a model is about matching the tool to the task. Every model makes different tradeoffs across three dimensions: speed, reasoning quality, and cost.
Speed
Faster models return results in seconds. Slower models may take 10–30 seconds for complex tasks. Speed matters most for:
- Tab completion (needs to be near-instant)
- Quick question-and-answer in chat
- Simple edits like renaming variables or reformatting code
For these tasks, a fast model keeps you in flow. Waiting 20 seconds for a variable rename suggestion breaks concentration.
Reasoning Quality
More capable models produce better results for:
- Complex algorithmic problems
- Architecting a system from scratch
- Debugging subtle issues where the root cause is non-obvious
- Explaining unfamiliar code deeply
- Multi-step refactoring across many files
For these tasks, the extra time and cost of a powerful model is worth it. A slightly wrong architecture suggestion from a fast model can cost hours to untangle.
Cost
Cursor charges usage in requests, not tokens. Different models consume different amounts of your monthly request budget.
- Lightweight models (cursor-fast, smaller models) use fewer or no credits
- Standard models like Claude Sonnet use one request per interaction
- Premium models like Claude Opus or o1 use more credits per request
This means using Claude Opus for every chat message will exhaust your budget faster than using Claude Sonnet.
Cursor Plans and Request Limits
Cursor has a free tier and paid plans, each with different monthly request allocations.
Free Plan
The free plan includes a limited number of premium model requests per month. After exhausting those, you can continue using lighter models or bring your own API keys. The free plan is enough to evaluate Cursor but limiting for daily professional use.
Pro Plan
The Pro plan includes significantly more premium model requests per month. This is what most individual developers who use Cursor daily will need. At this level, you can use Claude Sonnet generously without worrying about running out.
Business Plan
The Business plan adds team features, higher usage limits, and organizational billing. Teams that have adopted Cursor broadly typically use the Business plan.
What Counts as a Request
A request is consumed each time you send a message to the AI or trigger an inline edit. Long conversations, agent mode tasks that make many tool calls, and frequent use of premium models all count toward your monthly total.
If you run low on requests mid-month, Cursor will prompt you to upgrade or wait for the reset.
Using Your Own API Keys
If you exhaust Cursor's built-in quota, or if you want to use a model that is not in Cursor's default lineup, you can add your own API keys.
How to Add API Keys
Go to Cursor Settings → Models and enter API keys for the providers you want to use:
- Anthropic API key for Claude models
- OpenAI API key for GPT-4o and o1 models
- Google API key for Gemini models
When you use your own API key, usage is billed directly to your account with that provider. It bypasses Cursor's monthly request limits entirely.
When to Use Your Own Keys
Your own API keys make sense if:
- You are on a heavy-usage plan with a provider and want to route those calls through Cursor
- You want to use a newly released model that Cursor has not yet added to its own quota system
- Your organization already has enterprise agreements with AI providers
Many professional developers on the Pro plan start with Cursor's built-in quota and only add their own keys if they consistently hit limits.
A Practical Model Selection Strategy
Rather than trying to pick the perfect model for every situation, a simple strategy works well for most developers.
Default to Claude Sonnet
For the vast majority of coding tasks — writing functions, fixing bugs, answering questions, reviewing code — Claude Sonnet produces excellent results at a good speed. Make it your default and do not overthink it.
Upgrade to Claude Opus for Hard Problems
When you encounter a genuinely difficult problem — a subtle bug you cannot find, an algorithmic challenge, a complex refactoring decision — switch to Claude Opus. The better reasoning and more careful output is worth the extra wait and cost for these cases.
Use Reasoning Models for Algorithmic Tasks
If you are solving a problem that requires working through constraints or logic step by step — implementing a complex algorithm, designing a data structure, writing business logic with many edge cases — o1 or o3 may outperform other models noticeably. They are slow, but for this class of problem they are worth it.
Keep Fast Models for Completions
Do not overthink Tab completion. Cursor automatically selects a fast model for this. Leave it alone — the speed is what makes Tab completion useful.
Match Model to Context Window Needs
If you need to analyze a very large amount of code at once — reading an entire large codebase, comparing many files — a model with a large context window like Gemini 1.5 Pro can be useful. For most tasks, the standard context window of Claude Sonnet or GPT-4o is more than sufficient.
Summary
Cursor gives you access to multiple frontier AI models and lets you switch between them at any time in chat, inline editing, and agent mode. The key tradeoffs are speed, reasoning quality, and cost. For most coding work, Claude Sonnet is the right default. Upgrade to Claude Opus for complex problems and reasoning models for algorithmic challenges. Cursor's monthly request quota governs how many premium model calls you get, and you can bring your own API keys if you need more.
Key Takeaways
- Cursor supports Claude Sonnet and Opus, GPT-4o, o1/o3, Gemini, and more — all switchable from the model dropdown in chat or Cmd+K
- Claude Sonnet is the best default for most coding tasks: it balances quality and speed well
- Claude Opus and reasoning models like o1 are better for hard problems but are slower and use more of your monthly request budget
- Each Cursor plan comes with a monthly allocation of premium model requests; heavier models consume more credits per interaction
- You can add your own Anthropic, OpenAI, or Google API keys in Cursor Settings to bypass the built-in quota
- A simple strategy: default to Sonnet, upgrade to Opus for hard problems, and use reasoning models only for algorithmic or constraint-heavy tasks
Quiz
Discussion
Sign in to join the discussion.