Composer 2

Overview

Composer 2 is Cursor’s proprietary in-house coding model, representing significant advancement over Composer 1.5. Fine-tuned on Kimi K2.5’s mixture-of-experts architecture, Composer 2 combines exceptional coding performance with dramatic cost reduction, making it an attractive option for code generation and software engineering tasks.

Timeline

• 2026-03-21-AI-Digest - Composer 2 model announced with initial benchmark results
• 2026-03-23-AI-Digest - Performance evaluation and comparison against Opus 4.6
• 2026-03-24-AI-Digest - SWE-bench multilingual and extended benchmarking results
• 2026-03-31-AI-Digest - Model availability and pricing updates

Model Specifications

Architecture

• Base model - Kimi K2.5 (mixture-of-experts)
• Parameter configuration - 1T total / 32B active MoE
• Fine-tuning - Specialized for coding and software development tasks
• Developer - Cursor (in-house development)

Key Specs & Benchmarks

Coding Performance

• Terminal-Bench - 61.7 score (beats Opus 4.6’s 58.0)
• SWE-bench Multilingual - 73.7 score
• Practical advantage - Superior multi-lingual software engineering capability

Cost Efficiency

• Price reduction - 86% cheaper than Composer 1.5
• Value proposition - Dramatic cost improvement while maintaining or exceeding performance
• Target use case - High-volume code generation and refactoring tasks

Competitive Position

Composer 2 demonstrates that specialized fine-tuning of mixture-of-experts models can deliver both superior performance and lower costs compared to prior generations. The model’s strong Terminal-Bench score (exceeding Opus 4.6) and multilingual SWE-bench performance establish it as a competitive option for enterprise code generation workflows.

Ecosystem Integration

As Cursor’s proprietary model, Composer 2 is integrated into the Cursor IDE and available through the Cursor API for external integrations via MCP and similar protocols.