The Story

Anysphere, the company behind the AI code editor Cursor, has introduced Composer 2, a third-generation in-house model built to compete with leading AI coding systems.

According to the official announcement from Cursor, Composer 2 aims to deliver strong coding performance while significantly reducing cost per task compared to frontier models.

Performance highlights

• Benchmark results: Composer 2 scored 61.7% on Terminal-Bench 2.0, outperforming Anthropic’s Opus 4.6 at 58%.
• Close to frontier models: The model reportedly sits within a few points of GPT-5.4 on Cursor’s internal CursorBench evaluation.
• Rapid improvement: Internal benchmark scores have increased from 38% to over 61% across three generations since October.

Cost advantage

• Lower pricing: Composer 2 is priced at around $7.50 per million output tokens on its fast tier.
• Efficiency: This is roughly one-tenth the cost of GPT-5.4 and significantly cheaper than comparable frontier models.

What this signals

Cursor initially built its product by integrating top external AI models. With Composer 2, the company is moving toward owning its own model stack, giving it more control over cost, performance, and user experience.

Why it matters

The launch reflects a broader shift where application-layer companies are beginning to develop their own models rather than relying entirely on external providers.

If models like Composer 2 can approach frontier performance at significantly lower cost, they could reshape pricing dynamics in AI-powered development tools and influence how developers choose their coding assistants.