Introducing Composer 2

2026-03-19 · Source: Cursor Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Cursor has released Composer 2, a new frontier-level coding intelligence model, now available in Cursor. This model offers an optimal combination of intelligence and cost, priced at $0.50/M input and $2.50/M output tokens. Composer 2 demonstrates significant quality improvements across benchmarks, scoring 61.3 on CursorBench, 61.7 on Terminal-Bench 2.0, and 73.7 on SWE-bench Multilingual, surpassing its predecessor, Composer 1.5, which scored 44.2, 47.9, and 65.9 respectively. These advancements stem from continued pretraining and reinforcement learning on long-horizon coding tasks, enabling it to solve complex tasks requiring hundreds of actions. A faster variant with identical intelligence is also available at $1.50/M input and $7.50/M output tokens, becoming the default option.

Key takeaway

For AI Engineers and Software Developers seeking advanced coding assistance, Composer 2 presents a compelling option due to its enhanced performance on benchmarks like Terminal-Bench 2.0 and SWE-bench Multilingual, coupled with competitive pricing. You should consider integrating Composer 2 into your development workflow, especially for tasks requiring multi-step problem-solving, and evaluate the faster variant for latency-sensitive applications to optimize both cost and efficiency.

Key insights

Composer 2 offers frontier-level coding intelligence with improved performance and competitive pricing through continued pretraining and reinforcement learning.

Principles

• Continued pretraining strengthens base models.
• Reinforcement learning improves long-horizon task solving.

Method

The model's quality improvements derive from a continued pretraining run, establishing a stronger base for scaling reinforcement learning, which then trains on long-horizon coding tasks.

In practice

• Utilize Composer 2 for complex coding tasks.
• Choose the faster variant for speed-sensitive applications.

Topics

• Coding AI
• Large Language Models
• Reinforcement Learning
• AI Benchmarks
• Model Pricing

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Cursor takes on OpenAI and Anthropic with Composer 2, a code-only model built to match rivals at a fraction of the cost — Specialized, cost-effective AI models can compete with generalist leaders by focusing on narrow domains.
• Cursor’s new coding model Composer 2 is here: It beats Claude Opus 4.6 but still trails GPT-5.4 — Cursor's Composer 2 offers a cost-effective, deeply integrated AI coding model optimized for long-horizon agentic workflows.
• Cursor just beat EVERYONE. — Price-performance in coding AI is critical, with workhorse models like Composer 2.5 offering near-frontier capability at a fraction of the cost.
• Training Composer for longer horizons — Composer, a specialized model for agentic coding, has been trained for long-horizon tasks using a novel reinforcement learning process called "self-summarization," which integrates compaction directly…
• A technical report on Composer 2 — Continued pretraining and large-scale RL on realistic data significantly enhance coding model performance and efficiency.
• Composer 2.5 and I INTERVIEWED THE CEO OF ALPHABET — Composer 2.5 delivers near-frontier coding performance at a fraction of the cost, making it a leading workhorse model for most use cases.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Cursor Blog.