Many know Cursor as the code editor that integrates AI models from Anthropic and OpenAI. Now the startup is taking it a step further: on March 19, Cursor released Composer 2, its own AI model built specifically for code. And the opening statement is confident.
What Composer 2 Can Do
The model is trained exclusively on coding data. Not a generalist — a specialist. It handles prompts up to 200,000 tokens, works across multiple files, fixes bugs, and interacts with the command line.
On Cursor’s own benchmark (CursorBench), Composer 2 scores above 60 percent. According to Cursor, that puts it ahead of Claude Opus 4.6 but below GPT-5.4 in its highest configurations. On Terminal-Bench 2.0, which measures command-line capabilities, Composer 2 also outperforms Anthropic’s model.
The Price Is the Real Story
Here’s the kicker: Composer 2 costs $0.50 per million input tokens and $2.50 per million output tokens. That’s a fraction of what Claude Opus or GPT-5.4 cost. A faster variant is available at $1.50 and $7.50 respectively.
Cursor co-founder Aman Sanger explains the approach: because they trained solely on code, they could build a smaller model that’s significantly cheaper to run.
What This Means for the Market
The AI coding market is getting crowded. Claude Code, OpenAI’s Codex, GitHub Copilot, and now Cursor with its own model. The trend is clear: editor makers want to become independent from the big model providers.
Whether Cursor can sustain this long-term — against companies with significantly more resources — is an open question. But the pricing alone will turn heads. When a specialized model delivers comparable results at a tenth of the price, the conversation about the value of general-purpose models in coding is going to get interesting.
Sources: