Vib Coding Tool Cursor from Startup Any spareis Introducing Composerits first home, the proprietary Coding Large Language Model (LLM) as part of its Cursor 2.0 Platform Update.
Composer is designed to perform coding tasks quickly and accurately in a production-scale environment, representing a new step in AI-assisted programming. It’s already being used by Cursor’s own engineering staff in daily development—indicating maturity and stability.
According to Cursor, Composer completes most of the interactions Less than 30 seconds while maintaining a high level of reasoning ability in large and complex codebases.
The model is described as four times faster than similar intelligent systems and is trained for “agent” workflows.
Previously, cursors were supported "Web Coding" – Using AI to write code based on a user’s natural language instructions or complete code, even for someone untrained in development – Above other leading proprietary LLMS From the likes of Openai, Entropic, Google, and Zee. These options are still available to users.
Benchmark results
Benchmarking is done using Composer’s capabilities "cursor bench," An internal evaluation suite derived from real developer agent applications. A benchmark is a measure of not only accuracy, but also a model’s adherence to existing abstractions, style conventions, and engineering practices.
On this benchmark, Composer achieved frontier-level coding intelligence while generating 250 tokens per second -Twice as fast as fast inference models and four times faster than comparable frontier systems.
Cursor’s published comparison groups were grouped into several categories: “Best Open” (e.g., Kevin Coder, GLM 4.6), “Fast Frontier” (Hyko 4.5, Gemini Flash 2.5), “Frontier 7/2025” (the strongest available intermediate model), and “Best Frontier” (including GPT-5 and Killed Sonnet 4.5). Composer matches the intelligence of a mid-Frontier system while delivering the highest recorded generation speed of all tested classes.
A model built with a combination of reinforcement learning and expert architecture
Research scientist Sasha Rush K Curser provided insight into the development of the model Posts on social network xdescribing Composer as a Reinforcement Lord (RL) Mixture Expert (MOE) model:
“We used a large MOE model to train it to be really good at real-world coding, and very fast.”
Rush explained that the team co-designed both the Composer and Cursor environments to allow the model to run efficiently at production scale.
“Unlike other ML systems, you can’t abstract too much from a full-scale system. We designed the project and the cursor together to allow the agent to run at the necessary scale.”
Composer was trained on real software engineering tasks rather than static datasets. During training, the model ran within the full code base using a suite of production tools including file editing, semantic search, and terminal commands to solve complex engineering problems. Each training iteration involves solving a concrete challenge, such as developing a code modification, drafting a plan, or developing a goal specification.
The reinforcement loop improved both accuracy and efficiency. Composers learned to make effective instrument choices, use parallelism, and avoid unnecessary or speculative elements. Over time, the model developed emergent behaviors such as running unit tests, fixing linter errors, and executing multistep code autonomously.
This design enables Composer to work in the same runtime context as the end user, coupled with real-world coding situations.
From prototype to production
Development of the Composer followed earlier internal prototypes Cheetahwhich indicated low ratings for cursor coding tasks.
“The Cheetah was basically a V0 for high-speed testing of this model,” Rush said at X.
Cheetah’s success in reducing latency helped Cursor identify speed as a key factor in developer trust and availability.
Composer maintains this responsiveness while significantly improving reasoning and task generalization.
Developers using Cheetah during early testing noted that its speed changed the way they worked. One user commented that it was so fast that I could stay in a loop while working with it. ”
Composer maintains this speed but increases the capability for multi-step coding, refactoring, and testing tasks.
Integration with Cursor 2.0
Composer is fully integrated into Cursor 2.0, a major update to the company’s Agentic development environment.
The platform introduces a multi-agent interface, which allows Up to eight agents to run in parallel, Each in an isolated workspace using Git worktrees or remote machines.
Within this system, the composer can act as one or more of these agents, which can perform tasks independently or collaboratively. Developers can compare multiple results from concurrent agent runs and select the best output.
Cursor 2.0 also includes support features that enhance the effectiveness of Composer.
In-Editor Browser (GA) – Enables agents to run and test their code directly within the IDE, passing DOM information to the model.
Better code review -Multiple files have cumulative differences for rapid inspection of model-generated changes.
Sandboxed Terminals (GA) – Isolate commands from the agent-run shell for safe local execution.
Voice mode -Adds speech-to-text control to start or manage agent sessions.
While these platform updates expand the overall Cursor experience, Composer is positioned as the technical core that enables fast, reliable agent coding.
Infrastructure and training system
To train Composer at scale, Cursor combined PyTorch and Ray to build a custom reinforcement learning infrastructure for asynchronous training across thousands of NVIDIA GPUs.
The team developed a specialized MXFP8 MOE kernel and hybrid sharded data parallelizer, enabling large-scale model updates with minimal communication overhead.
This configuration allows the cursor to train low-precision models locally, without the need for post-training quantization, improving both inference speed and efficiency.
Composer’s training relied on millions of concurrent sandboxed environments—each a self-contained coding workspace—snapped in the cloud. The company adapted its background agent infrastructure to dynamically schedule these virtual machines, supporting the bursty nature of large RL runs.
Enterprise use
Composer’s performance improvements are supported by infrastructure-level changes to Cursor’s code intelligence stack.
The company has optimized its Language Server Protocols (LSPs) for faster evaluation and navigation, particularly in Python and TypeScript projects. These changes reduce latency when Composer interacts with large repositories or generates multi-file updates.
Enterprise users gain administrative control over Composer and other agents through team rules, audit logs, and sandbox enforcement. Cursor also supports the use of pooled models, SAML/OIDC authentication, and analytics to monitor agent performance across teams and enterprise tier organizations.
Pricing for individual users ranges from Free (Hobby) to Ultra (200/month), with extended usage limits for Pro+ and Ultra users.
Business pricing starts at $40 per month for teams, with enterprise contracts offering customized usage and compliance options.
Composer’s role in the evolving AI coding landscape
Composer’s speed, reinforcement learning, and focus on integration with direct coding workflows differentiate it from other AI development assistants such as GitHub’s Coplot or Doublet’s Agent.
Rather than serving as a passive suggestion engine, Composer is designed for continuous, agent-driven collaboration, where multiple autonomous systems interact directly with a project’s codebase.
This model levels specialization—training the AI to function in the real environment in which it will function. Composer is not trained solely on text data or static code, but within a dynamic IDE that mirrors production conditions.
Rush describes this approach as essential to achieving real-world reliability: the model learns not just how to generate code, but how to integrate, test, and improve it in context.
What this means for enterprise devs and vibe coding
With Composer, Cursor is introducing more than an agile model—it’s built to work within the same tools developers already rely on.
The combination of reinforcement learning, expert composition design, and tight product integration gives Composer a practical edge in speed and responsiveness that sets it apart from general-purpose language models.
While Cursor 2.0 provides the infrastructure for multi-agent collaboration, Composer is the primary innovation that enables these workflows.
It is the first coding model designed specifically for agent-based, production-level coding.