Composer 2 Technical Report

TL;DR

Composer 2 is a specialized AI model for software engineering that demonstrates advanced long-term planning, coding intelligence, and problem-solving abilities, trained through a two-phase process and evaluated on real-world benchmarks.

Contribution

It introduces a two-phase training process and a new benchmark for evaluating long-horizon coding tasks, advancing the state of domain-specific AI models.

Findings

Achieves 61.3 accuracy on CursorBench

Scores 61.7 on Terminal-Bench, 73.7 on SWE-bench

Demonstrates significant improvement over previous models

Abstract

Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learning to improve end-to-end coding performance through stronger reasoning, accurate multi-step execution, and coherence on long-horizon realistic coding problems. We develop infrastructure to support training in the same Cursor harness that is used by the deployed model, with equivalent tools and structure, and use environments that match real problems closely. To measure the ability of the model on increasingly difficult tasks, we introduce a benchmark derived from real software engineering problems in large codebases including our own. Composer 2 is a frontier-level coding model and demonstrates a process for training strong domain-specialized models. On our CursorBench evaluations the model achieves a major improvement in accuracy compared to previous Composer models (61.3). On public benchmarks the model scores 61.7 on Terminal-Bench and 73.7 on SWE-bench Multilingual in our harness, comparable to state-of-the-art systems.