Composer 1.5 - Cursor's self-developed AI programming model

What is Composer 1.5?

Composer 1.5 is Cursor's self-developed AI programming model, trained by scaling up reinforcement learning by 20 times. It's a "thinking" model, dynamically generating a reasoning codebase using thinking tokens, supporting rapid responses to simple questions and sustained deep thinking for complex tasks. Composer 1.5 has self-summarizing capabilities, automatically compressing its state and continuing exploration when context is exhausted. In internal benchmark tests, the model has surpassed its predecessor, showing significant improvements on challenging tasks.

Main features of Composer 1.5

• Intelligent coding proxy capabilityComposer 1.5 has proxy capabilities and can autonomously handle various coding tasks.
• Thinking-based reasoning mechanismDuring the query response process, the model generates a thinking token, which infers from the user's codebase and plans the next step.
• Adaptive thinking depthThe model is trained to answer simple questions quickly with minimal thought, and to think continuously on difficult questions, thus enabling flexible responses to tasks of varying difficulty.
• Self-summary functionTo handle longer-running tasks, Composer 1.5 has the ability to self-summarize, generating useful summaries when the available context is exhausted, allowing the model to continue exploring solutions.
• Recursive processing capabilityOn difficult examples, the self-summarization process is triggered recursively multiple times, allowing the model to maintain its original accuracy as the context length changes.
• Strengthen the scaling of learningComposer 1.5 is built by further expanding the scale of reinforcement learning by 20 times on the same pre-trained model. The computing power used for post-training even exceeds that of the pre-trained base model, bringing expected improvements in intelligence.

How to use Composer 1.5

• Obtain accessVisit the Cursor website and set Composer 1.5 as the default programming model in the settings.
• Select usage scenarioComposer 1.5 is particularly suitable for interactive coding scenarios, and the official recommendation is to prioritize the use of Composer 1.5 in daily development.
• Daily coding useSimply initiate a coding request in the Cursor editor. The model will automatically determine the task difficulty and adjust the depth of thinking, responding quickly to simple problems and automatically reasoning deeper into complex problems.
• Handling complex tasksFor challenging coding problems, Composer 1.5 automatically enters deep thinking mode, generating thinking tokens to reason about your codebase and plan action steps, without requiring manual setup.
• Long task processingWhen dealing with long-running tasks, Composer 1.5 automatically triggers a self-summary mechanism.

Application scenarios of Composer 1.5

• Complex code refactoringComposer 1.5 can understand the business logic of legacy code through deep reasoning, and automatically plan and safely execute large-scale refactoring tasks across multiple files.
• Long codebase maintenance With its self-summarizing mechanism, Composer 1.5 automatically compresses the context when dealing with large projects with tens of thousands of lines of code, continuously maintaining an understanding and consistency of the global logic of the codebase.
• Multi-file project developmentIn Agent mode, Composer 1.5 can autonomously analyze requirements, plan architecture, create and modify multiple files, and independently complete the development of complete functional modules.
• Troubleshooting and Bug FixesComposer 1.5 generates thinking tokens to deeply analyze the root causes of complex errors and deduce accurate fixes.
• Algorithm Design and Optimization When faced with complex algorithm problems, Composer 1.5 will continuously engage in in-depth thinking and explore multiple implementation paths to provide solutions with better time or space complexity.