Upgraded to 0.45.9, where is the Fusion Model

Discussion
1

Have upgraded to version 0.45.9, seems the Fusion Model is still not delivered.

This is the most wanted feature for me since the new tab model is announced via the blog.

Where is the Fusion Model and when will it come? Really cannot wait for this feature release.

2

What is “fusion model”?

3

See A New Tab Model | Cursor - The AI Code Editor

1 Like
4

Hey, we’ve rolled this out to all users now, so let us know how you get on!

1 Like
5

what about better codebase understanding is it also out ?

6

Not yet, we may have announced that too early!

1 Like
8

Hi, do you mean the new Tab Model has been already released and rolled out, but the better codebase understanding is still on the way?

9

It seems that the Fusion Model has a higher latency compared with the original one.

wechat_2025-02-05_180045_864
wechat_2025-02-05_180045_8642224Ă—942 318 KB

10

Hey, sometimes the latency can be higher if the demand for Tab is high, but we’re working to improve the capacity so this doesn’t happen in the future.

11

Hi, since the Tab Autocompletion is very latency sensitive for the developers, I am wondering whether we have some location specific servers for this tab autocompletion, for example in the Japan for the Asia developers?

12

Hey, the latency of Tab should already be pretty quick, are you seeing a long wait time for Tab suggestions to show?

13

Actually, I don’t have a very precise way to measure it.

When I check the logs, I often observe that the “streamingtime” frequently exceeds 1000, and in better cases, it’s around 500; the “actualTtftFromStart” values range between 300-600.

From practical usage, it’s acceptable for some time, but sometimes the wait time for code suggestions can be noticeable. The latency performance is not stable. Additionally, when I previously used SuperMaven, I felt its response time was excellent, with code suggestions appearing almost immediately after typing characters.

I know my description might not be accurate since I don’t have a scientific measurement method - it’s just based on my user experience. What I expect is that I shouldn’t need to pause during programming to wait for suggestions.

Again I am a big fan for the tab autocompletion. I don’t care about the model in agent or composer, but I can’t live without the tab autocompletion.

Even more, I can pay more for a very stable or dedicated auto completion functionality.

14

Hmm, thanks for reporting! Will throw this to the team to see if we can find an issue.