Cursor Launches Composer 2 AI Model: Input/Output Costs Slashed by 86% for Long-Horizon Coding
Agent: GLM-5 AI coding platform Cursor has unveiled Composer 2 and Composer 2 Fast, two proprietary models optimized for long-horizon coding tasks with a 200k context window. The release features aggressive pricing, cutting costs by up to 86% compared to its predecessor, while delivering competitive performance against models like Claude Opus 4.6.
AI coding platform Cursor announced on March 19 the release of two new programming models: Composer 2 and Composer 2 Fast. Designed specifically for "long-horizon agentic coding," these models feature a 200,000-token context window and are deeply optimized for complex, multi-step tasks within the Cursor environment.
A Shift Toward Long-Horizon Coding
Unlike traditional models that excel primarily at single-line code completions, Composer 2 focuses on "long-horizon coding," acting as an autonomous AI software engineer. Through specialized reinforcement learning, the model can handle complex tasks involving hundreds of operations. It is capable of reading codebases, determining modification strategies, editing multiple files, and running terminal commands seamlessly within the Cursor interface.
Crucially, Composer 2 is not a general-purpose standalone LLM. It is strictly locked into the Cursor ecosystem and does not offer an external API. Instead, it integrates deeply with Cursor's proprietary tool stack, including semantic search, browser control, and terminal operations.
Aggressive Pricing Strategy
The most notable aspect of the release is the dramatic reduction in cost. Compared to the Composer 1.5 released in February (priced at $3.50/$17.50), the new Composer 2 Standard is priced at just $0.50 for input and $2.50 for output per million tokens, representing an 86% decrease.
The Composer 2 Fast version is priced at $1.50 for input and $7.50 for output, roughly 57% cheaper than the previous generation. Additionally, Cursor has introduced a "cache read" discount to further lower daily operational costs for developers.
Performance Benchmarks
In third-party benchmark tests, Composer 2 demonstrated significant generational improvements. In the Terminal-Bench 2.0 test, which measures terminal operation capabilities, Composer 2 scored 61.7, outperforming Claude Opus 4.6 (58.0), though it still trails the industry benchmark GPT-5.4 (75.1).
Cursor emphasized a pragmatic "cost-performance" argument rather than claiming absolute superiority. The company positions Composer 2 as a solution that provides sufficiently excellent intelligent performance for daily programming tasks at a fraction of the cost of top-tier models.
