Cursor 0.50: New Tab model, Background Agent & Refreshed Inline Edit

Discussions  Featured Discussions
1

Hey!

Cursor 0.50 is now available to everyone. This is one of our biggest releases to date with a new Tab model, upgraded editing workflows, and a major preview feature: Background Agent

New Tab model

The Tab model has been upgraded. It now supports multi-file edits, refactors, and related code jumps. Completions are faster and more natural. We’ve also added syntax highlighting to suggestions.

Background Agent (Preview)

Background Agent is rolling out gradually in preview. It lets you run agents in parallel, remotely, and follow up or take over at any time. Great for tackling nits, small investigations, and PRs.

@ folders and full codebase context

You can now include entire folders in context using @folders. Enable “Full folder contents” in settings. If something can’t fit, you’ll see a pill icon in context view.

Refreshed Inline Edit (Cmd/Ctrl+K)
Inline Edit has a new UI and more options. You can now run full file edits (Cmd+Shift+Enter) or send selections directly to Agent (Cmd+L).

Faster agent edits for long files
Agents can now do scoped search-and-replace without loading full files. This speeds up edits significantly, starting with Anthropic models.

Multi-root workspaces

Add multiple folders to a workspace and Cursor will index all of them. Helpful for working across related repos or projects. .cursor/rules are now supported across folders.

image
image2048×1168 113 KB

Simpler, unified pricing

We’ve rolled out a unified request-based pricing system. Model usage is now based on requests, and Max Mode uses token-based pricing.

All usage is tracked in your dashboard

Max Mode for all top models

Max Mode is now available across all state-of-the-art models. It gives you access to longer context, tool use, and better reasoning using a clean token-based pricing structure. You can enable Max Mode from the model picker to see what’s supported.

More on Max Mode: docs.cursor.com/context/max-mode

Chat improvements

  • Export: You can now export chats to markdown file from the chat menu
  • Duplicate: Chats can now be duplicated from any message and will open in a new tab


MCP improvements

  • Run stdio from WSL and Remote SSH
  • Streamable HTTP support
  • Option to disable individual MCP tools in settings

Hope you’ll like these changes!

Full changelog here

13 Likes
2

cant wait for background agent to rollout for more users, hope i get it soon.

3 Likes
3

Definitely great improvements, the speed of AI responses and how fast it applies changes is remarkable. Also noted more accurate Tab model, didnt need to hint to it anymore what else related needs to be changed.

Will need time to check out all the other changes :slight_smile:

4

Many bugs as well, apparently…so many, I don’t know where to start… :frowning:

5 Likes
5

Maybe a skill issue, but I’m pretty sure I used background agents earlier and now I can’t seem to do it again.

1 Like
6

Hey @danperks does the “thinking” text count as tokens? E.g. Gemini produces incredibly amounts of thinking, are those part of the token charges, or is it just the actual input/response of the model?

2 Likes
7

I think this update is awesome!

@ericzakariasson Wondering: don’t you want to at least link to the new dashboard in navigation or account settings page? Currently its just deep linked from docs and here but otherwise wouldn’t be found

2 Likes
8

so many bugs. incorrect edits, attempting to edit the wrong files, context almost does not apply in many cases, looses track of tasks within the space of 1 prompt, applies multiple edits for a single error fix before getting it right, and the list goes on. I sure hope these are not being charged for with use based billing..

7 Likes
9

Can you outline what the difference of " Multi-root workspaces" vs normal VSCODE workspaces is? That screenshot seems misleading. If in folder “backend” would the path not be “.” ? and the workspace file should sit in root of “backend” and “frontend” ?

I have a similar structure at the moment and using two cursor instances because the backend is a different language and has a different complexity than frontend. But they share the same DB for example - I think, ideally it would be great to have one instance working with both (FE/BE). But I didn’t want to end up in issues with making too complex.

2 Likes
10

how to test bg agents

1 Like
11

Somehow, some directories cannot be found when using the @ syntax.

Cmd-L no longer references the current file.

Sometimes Cmd-L starts using some completely unrelated file as context.

Prompt following also appears less accurate compared to previous version.

Still missing special comments to tell internal linters to ignore a piece of code that mix multiple programming languages for code generation.

2 Likes
12

Hi,

Any tutorials on background agents will be appreciated.

Thanks

2 Likes
13

I’m looking forward to using the new invisible agent—it’s a great Easter egg! IT WORKS LIKE A CHARM! I’m just kidding, of course. I’m not sure why, but there’s an empty slot, and if I select it, it’s actually harder than to repone the menu.

Screenshot 2025-05-16 at 09.50.31
Screenshot 2025-05-16 at 09.50.31546×606 31.1 KB

2 Likes
14

What’s the link to the new dashboard? Or where is it documented? @arwed

1 Like
15

BG Agents are in review now (basically test users that try them out and help improving them)

16

The changelog has the link, I assume they still have to update all docs

Dashboard:

Changelog:

17

@Krad.Rocks Lets call it multi project workspaces, so basically you have a workspace where you can add mutliple projects (frontend/backend) in their separate folders and work on both, each keeps their own cursor rules that are specific to those projects. In the past Cursor supported only one project per workspace. Now you can index them and use them separately from same window/workspace.

@charles AI model providers (gemini, openai, anthropic) charge for thinking tokens as the machine has to generate them, process them again after thinking into final output.
they all charge for input tokens, thinking tokens, and output tokens. (some part as listed on cursor docu, like cached tokens for repeat sending of chat messages would be charged lower, depending on AI provider as well)

@cishiv I do not recall there ever being an background agent before :slight_smile:

18

I dont understand that every single update after stopping work …those guys from Cursor doesnt care or what just f…all and let others with huge problem … I think they got same as models … totally fools

1 Like
20

like anywhere

make a small as possible repeatable error and report that

1 Like
21

Same here! Backend in Python. Frontend in nextjs typescript. They of course share the same database. And I am curious to have this answered in more depth as well.

1 Like