Max Mode
Experience enhanced AI capabilities with Max Mode in Cursor
Max Mode gives you access to the full capabilities of Cursor’s advanced AI models. It’s designed for moments when you need additional processing power and deeper analysis.
Comparison
The main difference between normal mode and Max mode is context behaviour where Max mode is optimized to work through as much context as possible. In practice, this means:
- Larger context windows
- Up to 200 tool calls (without asking for continuation)
- Read file tool can read up to 750 lines
Context Window Comparison
Here’s what different context window sizes can handle in practice, with examples from real-world codebases:
| Tokens | Scale | Real-World Examples | What Fits |
|---|---|---|---|
| 10,000 | Small | Single utility libraries | A utility like Underscore.js, or a few React components |
| 60,000 | Medium | Utility collections | Most of a medium-sized library like Lodash |
| 120,000 | Large | Full libraries | Complete utility libraries or core parts of larger frameworks |
| 200,000 | Extra Large | Web frameworks | Full frameworks like Express, or runtime libraries like Tokio |
| 1,000,000 | Massive | Framework cores | Core of major frameworks like Django (without tests) |
Real-world token usage varies based on code style, comments, and tokenization. For example, Google’s documentation suggests that 1M tokens typically handles ~30,000-50,000 lines of code at 80 characters per line in practice.
Pricing
Max mode pricing is calculated based on tokens, with Cursor charging the model provider’s API price plus a 20% margin. This includes all tokens from your messages, code files, folders, tool calls, and any other context provided to the model. For detailed pricing, see models
Token usage from your account dashboard.
Token types
Model providers offer pricing in different tiers, depending on capabilities for caching etc. Here’s a breakdown of the various token types and what they mean. You can see an example of how tokens are determined using OpenAI’s tokenizer (note that this is just for demonstration - we use different tokenization under the hood).
| Type | Description | Example | Cost |
|---|---|---|---|
| Input | Tokens the model is seeing for the first time, | New messages | Expensive |
| Input (Cached) | Tokens the model has seen before and have decided to cache for future user | Previous messages in a chat | Cheap |
| Output | The returned tokens by the model | Model response, mostly code | Expensive |
How to use Max mode
To use Max mode,
- Open model picker
- Toggle
Max mode - Select a compatible model from the dropdown
FAQ
When should I use Max mode?
Max Mode works particularly well for the hardest problems where you need the model to think and reason the most. Normal mode is still recommended for most tasks.
Why don’t all models have Max mode?
Max mode is intended for the large, context intensive operations. Models with context windows of 100k tokens or less see diminishing returns with Max mode