Cursor launches in-house coding model to become independent from OpenAI, Anthropic
So far, it’s usually like this: When you use vibe-coding platforms like Lovable, Cognition, or Cursor, they mostly contain AI models from OpenAI, Anthropic, or Google inside – they ultimately serve as the engine for the performance of these tools. This also means that the providers are dependent on the model makers and transfer considerable sums to them for API usage.
Now, code editor Cursor has introduced its first in-house AI model for programming. According to the company, the model called “Composer” works four times faster than comparably powerful models. With this, the startup Anysphere behind Cursor frees itself from dependency on Anthropic and Co. Until now, Cursor has incorporated AI models from OpenAI, Anthropic, Google Gemini, and xAI.
Technical Characteristics of Composer
Composer was specifically developed for agent-based programming with low latency. According to Cursor, most work steps are completed in under 30 seconds. The model features various tools, including a codebase-wide semantic search, which should make working in large codebases easier.
According to Cursor, early testers of the model reported positive experiences with rapid iteration and showed confidence in the model’s ability to handle multi-step programming tasks.
New User Interface for Multi-Agent Workflow
Parallel to the release of Composer, Cursor is introducing a redesigned user interface called “Cursor 2.0.” According to the company, the new interface has been completely reconceptualized from the ground up and focuses on agent-based working instead of the traditional file-based view.
The new version allows multiple AI agents to work in parallel without interfering with each other. This is achieved through the use of Git worktrees or remote machines. Cursor states that having multiple models work on the same problem in parallel and then selecting the best result leads to improved outcomes, particularly for more complex tasks.
Additional Features
In addition to the basic architecture, Cursor has identified two other areas that are relevant when using AI agents for programming: code review and testing changes.
Cursor 2.0 therefore includes improved features for quickly reviewing changes made by agents. Additionally, a native browser tool has been integrated that allows Cursor to test its own changes and iteratively improve them until a correct final result is achieved. Users can still switch back to the classic IDE view if needed or open files in the new layout.
What is Magic Doing?
It will now be interesting to see when the startup Magic with roots in Austria finally enters the arena. Founders Eric Steinberger and Sebastian De Ro have raised more than half a billion dollars to develop their own AI model specialized in coding. This model is not yet on the market.
