Web & Javascript: Learning by Building with Cursor AI

Composer 2 - The greatest LLM model ever announced by Cursor

I have some incredible news to share.

Composer 2 is now available in Cursor.

Coding performance is at a cutting-edge level, and the price is 0.50 per 1 million input tokens and 2.50 per 1 million output tokens,

It offers a new optimal combination of intelligence and pricing.

Frontier-level coding intelligence

It scored excellently across all benchmarks.

Here is a description of each benchmark.

CursorBench

This is a test created by Cursor themselves. You can think of it as them testing "how well it actually works in our tool." For example, they give the AI real development tasks like "Fix this bug" or "Add this feature" and score how well it resolves them. There is a limitation that it could be designed somewhat in their favor since it is based on their own standards.

Terminal-Bench 2.0

It tests how well the AI performs in the terminal (typing commands on a black screen). For example, it gives tasks that require performing multiple steps in sequence, such as "set up the server, install packages, run tests, and check the results." Since these situations require executing multiple commands consecutively rather than finishing with a single click, it evaluates the AI's "ability to carry out tasks independently." It is reliable because it allows for comparison against coding agents like Claude Code or Codex using the same criteria.

SWE-bench Multilingual

It is one of the most prestigious coding benchmarks in the AI field. It takes GitHub issues (bug reports) from real open-source projects and tasks the AI with solving the problem by directly modifying the code. For example, if you provide a prompt like "Fix this error occurring in this library," the AI analyzes the code and generates a fix patch. The "Multilingual" tag means it includes various languages such as Java, TypeScript, and Go, in addition to Python. Since it is an objective test managed by an external organization, it is the most frequently cited benchmark in the industry.

Available on Pro plans and above.

For developers who develop using only Tab, the $20 plan is sufficient.

Try Composer 2

The price for Composer 2 is 0.50 per 1 million input tokens and 2.50 per 1 million output tokens.

There is also a faster version that offers the same level of intelligence.

We will be providing this faster version as the default option. For more details, please check the model docs.