What is 3.7 Sonnet MAX?

Discussion
1

In the newest 0.47.5 update, changelog mentions

  • 0.47.5: Client-side support for upcoming 3.7 Sonnet MAX

Any information on what that is? I don’t seem to find any specific information when searching around.

2

I think more information on this will be coming out as the actual model is released.

My understanding is that it’s an enhanced version of 3.7 Sonnet. It has enhanced capabilities by using more compute and taking longer to think.

1 Like
3

presume it will be for the “max” thinking toggle

4

when?

5

I think very soon.

6

I think more likely its building on already existing Claude features that Cursor didn’t previously take advantage of, foremost the already confirmed enterprise 500k token large context model: Claude for Enterprise \ Anthropic

There were also some advanced (higher cost) new tool calling capabilities in Claude 3.7 that Cursor hadn’t implemented yet such as Claude Code (internal agentic AI), long running tasks and such:

In the coming weeks, we plan to continually improve it based on our usage: enhancing tool call reliability, adding support for long-running commands, improved in-app rendering, and expanding Claude’s own understanding of its capabilities.
Claude 3.7 Sonnet and Claude Code \ Anthropic

7

image
image469×294 39.7 KB

Cursor 0.47.7 is out with MAX mode! (But it comes at an extra cost) https://www.cursor.com/changelog

8

I was partially correct it seems, larger context window, improved tool calling 🚀 Claude 3.7 Max - Out Now!

9

Yea, the extra cost part is where I’m now stuck.

Is this task worth the extra costs, or would regular 3.7 do fine too for no cost?

Now I will end up with decision fatigue before ever starting :rofl:

10

For those of us who haven’t had a reason to use usage-based pricing, it would be nice to have more clarity on what constitutes “tool use”, which isn’t fully clear from the docs.

Tools are only mentioned in the Agent section, but it also says that “reading & writing code” and “codebase searching” are considered tool use.

Is every single edit considered multiple tool uses since Cursor obviously has to search and read the code before writing?

e.g. If Agent makes 5 small inline edits, that might be billed as 10-15 “tool users” + once for the prompt?

Where I’m confused is Edit mode makes no mention of tool use in the docs, but a single request could make the identical edits as above.

For now I’ll stick with Edit mode, but more clarity could prevent users from racking up expensive invoices unexpectedly.