Cursor Composer 2: Frontier-Level Coding at Practical Pricing
Cursor has released Composer 2, its latest in-house coding model, boasting frontier-level performance, improved long-horizon task handling, and lower cost than competing fast models.
Why it matters
Composer 2's combination of frontier-level coding capabilities and aggressive pricing makes it a significant development in the AI-assisted coding landscape.
Key Points
- 1Composer 2 delivers large improvements on coding benchmarks like Terminal-Bench 2.0 and SWE-bench Multilingual
- 2Cursor's first continued pretraining run provides a stronger base for the model
- 3Reinforcement learning is used to optimize Composer 2 for long-horizon coding tasks
- 4Cursor also introduces a faster Composer 2 variant with the same intelligence
Details
Composer 2 is Cursor's latest in-house coding model, positioned as a frontier-level and cost-effective option for agentic software work. The key improvements include stronger coding performance across multiple benchmarks, continued pretraining to build a better foundation, and reinforcement learning to handle long-horizon coding tasks. Cursor claims Composer 2 can solve challenging tasks requiring hundreds of actions, indicating optimization for sustained multi-step software development. Additionally, Cursor is offering a faster variant of Composer 2 with the same underlying intelligence, providing a practical option for everyday use.
No comments yet
Be the first to comment