s AI coding tools like Anthropic's Claude Code and OpenAI's Codex soar in popularity, Cursor is raising its game.

On Thursday, Cursor launched Composer 2, the AI startup’s most capable coding model yet. While there are many AI coding models on the market from leading firms, Composer 2’s appeal lies in its ability to perform comparably to those models at a much lower cost.

“Developers need models good enough to trust and cheap enough to use constantly,” Cursor told The Deep View. “At enterprise scale, cost per token directly affects deployment economics.”

For instance, on the Terminal-Bench 2.0, a benchmark made to evaluate real AI agent workflows, Composer 2 performed comparably to Anthropic's Opus 4.6 and trailed OpenAI's GPT-5.4, while remaining the lowest-priced option by a significant margin.

Quick look below:

• Composer 2: Score: 59.8%, Price: $0.50/M input and $2.50/M output
• Opus 4.6: Score: 58.0%, Price: $5/M input and $25/M
• GPT-5.4: Score: 75.1%, Price: $2.50/M input and Output: $15.00/M output

Composer 2 also outperforms all preceding models on the CursorBench, Terminal Bench 2.0 and SWE-Bench Multilingual.

Composer was trained for long-horizon tasks using reinforcement learning, making it capable of solving challenging coding workloads. Composer 2 achieved improved model quality and lower cost through continued pretraining, enabling Cursor to scale reinforcement learning, according to the company's blog post.

However, Cursor notes that beyond the model's prowess, what sets the platform apart from competitors is its full development environment.

“Our users can choose from Claude, GPT, Gemini, and our own Composer models in one environment,” said Cursor. “The key is product scope: we're a full development environment with long-running agents, code review, a plugin marketplace, automations, and deep enterprise security."

Our Deeper View

As the AI coding tool race heats up, developers gain access to increasingly capable development tools. This is significant because it not only enables experienced developers to produce what they need faster, but also allows people with ideas and no technical background to throw their hat in the ring. The outcome in the near future should be a new wave of software and application innovation that solves real-world problems and needs. And it's not just coming from Cursor. Google’s latest AI studio upgrades the entire vibe coding experience, now allowing users to build multi-player experiences, add databases and authentication, connect to real-world services, access a more powerful agent, and more. There's never been a better time to start building your own stuff.