Skip to main content

Cursor's New AI Coding Assistant Rivals Top Models at a Fraction of the Cost

Cursor Levels the Playing Field for AI-Assisted Coding

The AI programming landscape just got more interesting. Cursor, known for its intelligent code editor, has released Composer 2.5 - a powerful new model that punches above its weight class while keeping costs surprisingly low.

Image

What makes this release stand out? The team didn't just tweak their existing model - they completely overhauled the training approach. Using Moonshot AI's open-source Kimi K2.5 as a foundation, they pumped in 25 times more synthetic task data than previous versions. Even more impressive? They dedicated 85% of their computing budget solely to reinforcement learning and extended training.

Performance That Turns Heads

The results speak for themselves. In industry-standard SWE-Bench multi-language tests, Composer 2.5 scored an impressive 79.8%. It followed up with a 63.2% score on CursorBench v3.1 evaluations - numbers that put it in the same league as premium models from Anthropic and OpenAI.

But here's where it gets really interesting: while the performance matches high-end competitors, the price doesn't. Composer 2.5 operates at just $0.50 per million input tokens and $2.50 per million output tokens - a fraction of what other major players charge.

Options for Every Developer

Cursor understands that different developers have different needs:

  • Standard version: Balances cost and performance perfectly for most users
  • Turbo variant: Delivers even faster responses while maintaining quality (priced at $3/$15 per million tokens)

Both options are now live in the Cursor editor, ready to help developers write better code faster.

Building Tomorrow's AI Today

The company isn't resting on its laurels either. In partnership with SpaceX and xAI, Cursor is already training next-generation models on the massive Colossus-2 cluster - a system packing millions of H100 GPUs with ten times the computing power of their current setup.

Key Points:

  • Performance breakthrough: Matches top-tier models like GPT-5.5 in coding tasks
  • Cost advantage: Pricing structure makes advanced AI assistance accessible to more developers
  • Future-ready: Next-gen models already in development with major tech partners