Cursor announced Composer 2 on March 19, 2026 — a code-only model built for multi-file edits, refactoring, and long-running coding tasks that beats Claude Opus 4.6 on most programming benchmarks.
The model scores 61.3 on Cursor's internal CursorBench, a major improvement over Composer 1.5 (44.2) and competitive with Claude Opus 4.6 (58.2) and GPT-5.4 Thinking (63.9). On Terminal-Bench 2.0, Composer 2 outperforms both GPT-5.4's low configuration and Claude Opus 4.6.
Composer 2 supports prompts with up to 200,000 tokens and can generate code, fix bugs, and interact with command line interfaces. It's designed specifically for the coding domain rather than general-purpose conversation.
Pricing is dramatically lower than frontier models: $0.50/M input and $2.50/M output tokens for the standard tier, with a faster variant at $1.50/M input and $7.50/M output. This represents an 85% cost reduction from Composer 1.5.
The technical foundation is notable: Composer 2 is built on the Chinese open-source model Kimi K2.5, with roughly a quarter of pretraining from the base model while Cursor did the rest through fine-tuning and continued training.
The release positions Cursor as not just an IDE but a model provider — directly competing with OpenAI and Anthropic on coding-specific workloads at significantly lower price points.