A new AI coding model from startup Cursor delivers performance comparable to industry leaders at one-tenth the cost, challenging OpenAI and Anthropic's dominance in developer tools.
Composer 2 launched on March 19 through Cursor's popular AI code editor, offering developers access to advanced coding assistance for $0.50 per million input tokens and $2.50 per million output tokens. A faster variant costs $1.50 per million input tokens with $7.50 per million output tokens.
The pricing undercuts established competitors by a wide margin while maintaining competitive benchmarks on coding tasks. According to internal testing using CursorBench, Composer 2 achieved scores above 60%, placing it ahead of Claude Opus 4.6 but behind GPT-5.4's high and medium configurations.
Cursor's approach focuses on specialized training rather than general-purpose capabilities, with Bloomberg reporting the model was trained solely on coding datasets to optimize cost efficiency for development work.
San Francisco-based Cursor has grown rapidly since launching its first AI coding assistant in 2023, now serving more than one million daily active users including enterprise clients like Stripe and Figma.
The company secured a $29.3 billion valuation last November and is reportedly raising additional funding that could push its valuation to approximately $50 billion.
Composer 2 integrates directly with Cursor's agent workflow system, providing access to semantic code search, file operations, shell commands, browser control, and web access capabilities within the development environment.
Developers using the platform can choose between multiple pricing tiers: a free Hobby plan, Pro at $20 monthly, Pro+ at $60 monthly, or Ultra at $200 monthly for individual users.
Business teams pay $40 per user monthly for Teams access, while Enterprise customers receive custom pricing with pooled usage, centralized billing, privacy controls, single sign-on integration, audit logs, and administrative tools.
The launch comes as competition intensifies in AI-assisted development tools, with users increasingly comparing offerings across platforms including Anthropic's Claude Code for terminal-first workflows.
