Cursor Launches Composer 2: Custom Coding Model with Top Performance

Cursor, the AI-powered code editor that became an industry reference, just released Composer 2 — an AI model trained exclusively for code, with performance surpassing established competitors.

What is Composer 2?

Unlike generic models like GPT or Claude, Composer 2 was trained specifically for programming tasks. It deeply understands code contexts, project structures, and engineering best practices.

Impressive benchmarks:

• Terminal-Bench 2.0: frontier-level performance on terminal tasks
• SWE-bench Multilingual: outperforms previous versions across all languages
• Complex tasks with hundreds of sequential actions solved accurately

Revolutionary Pricing

The price is what stands out: $0.50/M input tokens and $2.50/M output tokens — that's 86% cheaper than the previous version, with superior performance.

In practice:

• A coding session that cost $1 now costs pennies
• Teams can use AI generously without blowing budgets
• Individual developers get frontier-quality AI at an accessible price

Competitive Comparison

According to independent benchmarks:

• Vs Claude Opus 4.6: Composer 2 wins on most coding tests
• Vs GPT-5.4: close on general tasks, surpasses on code-specific tasks

Cursor is proving that the future of AI-assisted programming doesn't depend on generalist models — specialized coding models are the way forward.