Interactive agentic coding
Iterate quickly through a write, run, inspect and fix loop without waiting on a slower reasoning model for routine work.
Independent tool overview
Cursor Composer at a glance
Composer 2.5 is Cursor's current fast, cost-focused agentic coding model for searching repositories, editing files, running commands and completing long software tasks.
Visit the official Cursor Composer site ↗
- Current version
- Composer 2.5
- Provider
- Cursor / Anysphere
- Context window
- 200,000 tokens
- Model ID
- composer-2.5
- Availability
- Cursor products and Cursor Agent SDK
- Last reviewed
- August 30, 2026
Overview
What Cursor Composer is
The Composer model introduced with Cursor 2.0 in October 2025 has been superseded by Composer 2.5. The current version has a 200,000-token context window and is tuned for long-horizon coding, tool selection, file edits and terminal operations inside Cursor's local, cloud, web, CLI, mobile and SDK agent surfaces. Older Composer 2 model IDs are routed to 2.5 for compatibility.
Composer's value proposition is the combination of capable coding and fast iteration at relatively low token prices. It can search a repository, read and edit files, run commands, browse and use connected tools through Cursor's agent harness. It still needs the same engineering controls as any coding agent: a narrow brief, tests, secret protection, diff review and explicit approval before destructive commands, deployments or changes to production data.
Use cases
Who Cursor Composer is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Repository-wide changes
Search a large codebase, trace related files and apply coordinated edits across multiple modules.
Long-running implementation tasks
Handle multi-step work that needs repeated tool calls, test runs and corrections over an extended session.
Cloud-agent delegation
Run work against connected repositories in Cursor's managed environment and return branches, pull requests or demos for review.
Programmatic coding agents
Build local or cloud agents with the Cursor SDK while selecting Composer 2.5 explicitly.
Capabilities
Core Cursor Composer features
Long-horizon coding
Composer 2.5 is trained and tuned to sustain multi-step software tasks with many tool actions.
Repository search
Searches files, folders and text patterns to assemble relevant code context.
File editing
Creates and applies changes across source, configuration, tests and documentation.
Terminal execution
Runs commands, follows process output and uses the results to continue debugging or implementation.
Browser verification
Can operate a browser for screenshots and visual checks when used through supported Cursor agent tools.
Web and external tools
Can search the web and connect to approved MCP servers or Cursor Marketplace plugins.
Standard and Fast variants
Offers a lower-cost standard tier and a higher-throughput Fast tier with the same model positioning.
Local and cloud runtimes
Works in a local repository or a durable Cursor cloud agent connected to source control.
Cursor Agent SDK
Public-beta TypeScript and Python SDKs expose agent lifecycle, streaming, cancellation and local or cloud execution.
Privacy Mode
Prevents code from being used for training; team admins can enforce it across members.
Process
How the Cursor Composer workflow works
Step 1
Define the boundary
Specify the desired behavior, allowed files, non-goals, acceptance tests and actions that require approval.
Step 2
Provide repository guidance
Point to project rules, architecture notes and canonical commands while excluding secrets and irrelevant generated files.
Step 3
Ask for inspection first
Have Composer identify the current path, dependencies and likely change surface before applying a broad edit.
Step 4
Implement in small checkpoints
Prefer reviewable steps with focused tests over one unbounded request that touches the entire repository.
Step 5
Verify independently
Run formatting, type checks, tests and browser or runtime checks, then inspect the actual diff and generated artifacts.
Step 6
Approve consequential actions
Keep merges, deployments, migrations, credential changes and production writes under explicit human control.
Cost
Cursor Composer pricing and free plan
Composer 2.5 consumes Cursor's first-party model usage pool on eligible plans, with additional use billed by tokens. The lower-cost standard variant is $0.50 per million input tokens and $2.50 per million output tokens; Fast is the product default and costs $3 and $15 respectively. Cache-read rates are lower.
Composer 2.5 Standard
$0.50/M input · $0.20/M cached input · $2.50/M output
Token-based standard variant optimized for lower cost per task.
- Available through Cursor's agent surfaces and SDK
- Draws from eligible included usage before on-demand billing
Composer 2.5 Fast
$3/M input · $0.50/M cached input · $15/M output
Higher-throughput variant and the default Composer experience in the product.
- Six times the standard input and output token rates
- Use when lower latency justifies the premium
Cursor Hobby
$0
Free editor plan with limited Agent requests and access to Composer.
- No credit card required
- Best for evaluating workflow fit rather than sustained agent usage
Cursor Individual
$20, $60 or $200/month
Pro, Pro+ and Ultra plans with progressively larger Agent usage pools.
- Pro: $20/month
- Pro+: $60/month
- Ultra: $200/month
- Additional on-demand usage can be purchased
Cursor Teams
$40 or $120/user/month
Standard and Premium team plans with administration and enforced privacy controls.
- Teams Standard: $40/user/month
- Teams Premium: $120/user/month
- Enterprise is custom
Pricing checked . Check current pricing at the source ↗
Assessment
Cursor Composer strengths and limitations
Where it stands out
- Fast interactive loop for routine and moderately complex coding work
- Standard variant has low published input and output token rates
- Designed around repository search, editing, terminal use and long agent trajectories
- Available across Cursor's desktop, web, CLI, mobile, cloud and SDK surfaces
- 200,000-token context supports substantial repository and task state
- Local and cloud execution cover hands-on pairing and delegated work
- SDK enables durable programmatic agents rather than only editor chat
- Privacy Mode can be enabled by individuals and enforced for teams
What to consider
- The original Composer announcement and benchmark results describe an older model generation; current users receive Composer 2.5
- Composer 2.5 is available only through Cursor's products and agent SDK, not as a general-purpose raw chat-completions endpoint
- Fast is the default variant but costs six times the standard input and output token rates
- A 200,000-token window does not guarantee that the model retrieves or prioritizes the right repository context
- Agentic edits can introduce regressions, insecure defaults, unnecessary dependencies or partially completed migrations
- Terminal and browser tools increase the impact of a mistaken or malicious instruction
- Cloud agents require Cursor to retain repository access and code for the duration of their work
- AI requests route through Cursor's backend even when a user supplies a separate provider key
- Privacy Mode prevents training use but does not mean code never leaves the local machine
- Cursor's security page says extension signature verification is disabled and VS Code Workspace Trust is disabled by default
- The Cursor Agent SDK is in public beta and can change
- Vendor-published benchmark gains should not replace testing on the team's own repositories and acceptance criteria
- Included model usage varies by subscription and actual token consumption
Compare
Cursor Composer alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
Codex
OpenAI's broader agentic coding environment for local and cloud tasks, long-running work and programmatic workflows.
Explore Codex →Coding
Claude Code
A terminal-centered coding agent that is a strong fit for deep repository reasoning and developer-controlled workflows.
Explore Claude Code →Coding
Cursor Router
Cursor's automatic model selection when cost, intelligence or reliability matters more than pinning Composer.
Explore Cursor Router →Coding
Devin Desktop (formerly Windsurf)
A competing agentic development environment now integrated with Devin's broader software-engineering platform.
Explore Devin Desktop (formerly Windsurf) →Questions
Cursor Composer FAQs
What is Cursor Composer?
Composer is Cursor's first-party agentic coding model family. The current version is Composer 2.5, which searches repositories, edits files, runs commands and works through Cursor's agent runtime.
Is the original Composer model still current?
No. The October 2025 announcement introduced Composer 1. Cursor later released Composer 2 and 2.5; older Composer 2 identifiers now route to 2.5.
How much does Composer 2.5 cost?
Standard costs $0.50 per million input tokens, $0.20 per million cached input tokens and $2.50 per million output tokens. Fast costs $3, $0.50 and $15 respectively. Eligible Cursor plans include usage pools before on-demand billing.
Is Composer available on Cursor's free plan?
Yes. Cursor's Hobby plan lists access to Composer with limited Agent requests. Paid individual plans start at $20 per month.
Can I use Composer outside the Cursor editor?
Yes, through Cursor's web, CLI, mobile, cloud-agent and public-beta Agent SDK surfaces. Cursor does not document Composer as a standalone raw model-inference API.
What is Composer's context window?
The current official model page lists a 200,000-token context window for Composer 2.5.
Does Cursor train on my code?
With Privacy Mode enabled, Cursor says code is not used for training by Cursor or model providers. AI requests still send selected prompts and code context through Cursor's backend and inference infrastructure.
Should Composer be allowed to merge or deploy automatically?
Only after the team has established strong tests, scoped credentials, approvals and rollback. Start with human-reviewed branches and treat agent output like code from an untrusted contributor.
Bottom line
Our Cursor Composer verdict
Composer 2.5 is a compelling default inside Cursor when developers want a quick agent loop and predictable token economics without jumping to the most expensive frontier model. Its advantage comes from the surrounding search, editing and execution harness as much as the model. Teams should compare it on real tasks and keep production authority behind tests and review.
Visit Cursor Composer website ↗
.png)
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.