What's new in Cursor 2.0?

29 Oct, 2025

Cursor has released v2.0 today with pretty interesting features.

Here are some highlights

Voice Mode

Built-in speech-to-text conversion 🎤

In the early version, the browser is more of headless. But now they’ve preview, devtools, etc.

You can select a element and chat with it.

IDE like Windsurf, Trae, etc has this feature for quite sometime. But it’s good support for this in cursor itself.

They’ve improved their code review UI.

Agent commands runs in a sandbox (on macOS)

In macOS, it uses sandbox-exec (similar to how it is handled in Claude Code or Codex)

Shell commands that are not already in allow list will automatically run with read/write access to your workspace and no internet access

They’ve seperate doc about it if you’re curious about it.

Recently you might have noticed Cheetah model that was performing good enough but super fast.

This is the next iteration of that model.

You can run a single prompt across 8 different AI model in parallel.

This might be useful when you need more accuracy and ready to burn some money 🤑

Run Plan model with one model and build mode with another model.
Improved prompt UI - They’ve removed explict context declaration like @Web, @Recent Changes, @Link, @Linter Errors, etc. And claim that agents can self-gather the context without needing to be explict.
Sharable team commands
Performance improvements 😅