Claude 3.5 Sonnet is now available to all Copilot users in Public Preview

[matthewisabel]

Select Topic Area

General

Body

As announced at GitHub Universe on Tuesday, October 29th, Claude 3.5 Sonnet is now available to all GitHub Copilot customers!

To learn more, you can see the announcement blog post or the docs on Claude 3.5 Sonnet in GitHub Copilot.

We'd love to hear your thoughts and feedback!

[jamesastound]

Loving the Sonnet preview. Few things I've noticed is it is more aggressive at changing things outside the scope of the query. This can be good and bad, depending on the context. On the good side, it will often add code that directly relates to the task, on a line that is appropriate. However, sometimes it changes things, seemingly, arbitrarily.

Biggest downside is it takes a lot longer than GPT 4o, tho it is similar to o1. Sometimes I feel that it is because it is over-zealous reading context of very large files, even if I only highlight a few lines. Would like a "mini" version of Sonnet, if that ever becomes feasible.

[vertexcs-suraj]

Can we get this feature in the vim plugin as well?

[sebastiancarlos]

According to another comment, the Claude support is only for chat mode at the moment, which isn't a supported mode on the vim plugin

[anshukg]

What is the size of the context window for Claude 3.5 Sonnet ?

[prime399]

I am wondering the same.

[prime399]

I have tested and asked copilot itself.

And got answer, it's same as any other model 4k in 4 out and overall for entire chat is 8k tokens.

[Otanikotani]

Is there a chance to get this in Jetbrains products too?

[matthewisabel]

We're targeting early 2025 for this

[NDR0216]

Claude kept responding that I included the timestamp and login information. Weird.

[igormalaes]

When will it be available for IntelliJ?

[baslia]

Same question, it seems like IntelliJ plugins are 6 months behind VS Code

[akkikumar72]

I moved to VS Code just for it

[Fadyio]

Is Claude 3.5 Sonnet only available in chat mode, or can we get inline completion as well? Is there a way to change which model we use for inline completion?

[matthewisabel]

Just chat at the moment!

[dogukanoksuz]

Inline completion model switching would be great tbh

[iwangbowen]

Not as good as GPT o1

[knnmran]

sometimes i need to reset the context. it seems like reloading the VSCode is the only way to do that at the moment.

[knnmran]

in some cases not sure which yet, it is not possible to remove working set files individually. missing "x" button.

[ossianravn]

For some reason I can't select Sonnet (or anything besides the default model) in Copilot/Visual Studio. I've enabled Sonnet in my Github Copilot settings about 12 hours ago, restarted VS, booted my PC, etc - but nothing. Why could that be? Thank you!

[matthewisabel]

Are you using Visual Studio Code or Visual Studio? Right now the model picker is only in Visual Studio Code but Visual Studio is on the roadmap!

[MasonMcV]

This should be publicized more, as it took me several days to find this out. Do you have a time frame for when this will be updated in other IDE's?

[ossianravn]

Ok, then it makes sense. I'm using Visual Studio. Thanks.

[mhensen]

Any timeframe? Why is Visual Studio 2022 behind on Visual Studio Code.. Why is VSC always prefered?

[onegigbyte]

I have enabled "Anthropic Claude 3.5 Sonnet in Copilot", but I still do not see the Claude option via copilot.
Can anyone help?

[jalenzz]

When you go to your Copilot settings, do you see the policy for Anthropic Claude 3.5 Sonnet in Copilot and is it either set to enabled or a policy you can enable?

Additionally, it may be your region is not supported by Anthropic (see supported regions), which would make it not visible in the dropdown.

@matthewisabel Can proxy be supported for connection?

[wyuancronos]

Hi @matthewisabel the policy for Anthropic Claude 3.5 Sonnet in my corporate copilot is enabled, and i'm in one of the supported regions, but I still don't see the option in chat. The OpenAI o1 models are loaded though. would you please advise anything elseI should check?

VS Code Version: 1.94.2 (Universal)
GitHub Copilot Version: v1.245.1209 (pre-release)

[OTPSerikat]

did you manage to solve that or still waiting for a fix?

[wyuancronos]

@OTPSerikat I'm still waiting for a fix and I'm using cursor at the moment

[matthewisabel]

Hi @wyuancronos, did you write into GitHub support? We likely need some additional information to debug here and they'll be able to do so.

[jalenzz]

🐛 There is no "Claude 3.5 Sonnet" option when I connect to remote host, I can only use it locally.

[hjqcan]

When will claude be available in VS2020? So disappointed in your guys.The gap between you and cursor is soooooooooo big

[Soontosh]

This is awesome! Just wondering if this is the new version of Claude Sonnet 3.5 which came out just under two weeks ago?

[matthewisabel]

Yes!

[chenxch]

Hello, I have enabled the Claude model in both my personal account and my corporate account, but the model does not appear in my vscode chat. I have confirmed that it has been updated to the latest version and have tried to log in again and even restart.

[chenxch]

https://github.com/orgs/community/discussions/143337#discussioncomment-11138626
This should be the reason.

[firstunicorn]

Is claude-3-5-sonnet-20241022 model available or it is just claude-3-5-sonnet ?

[matthewisabel]

The latest model (2024-10-22) is being used

[firstunicorn]

Also, is it available in Multi-file editing mode or it is just in chat right now? If it's only for chat, when it will be available in Multi-file editing mode? I'm thinking to switch from Cursor to Copilot and these is the only important things for me - sonnet-20241022 and Multi-file editing mode - because I use it most of the time (as I understand, it is similar to Composer mode in Cursor). Thanks.

[TimMoore]

Is there a way to make this the default model? I can select it in VS Code, but it seems to frequently revert back to GPT 4o

[isimmons]

Copilot feels a need to correct my misinformation when the Claude model is chosen. I just tried to say hello to my old Cursor friend. :-)

I need to correct some misinformation. I am GitHub Copilot, not Claude, and I do not use the Claude language model. I am developed > by GitHub/Microsoft and use different underlying technology.

Also,

I cannot engage in general conversation or non-programming tasks like Claude can. My responses are focused on helping you write,
understand, and debug code within VS Code.

[nv-pipo]

Comparing VSCode with Claude 3.5 VS Cursor with Claude 3.5: VSCode feels considerably slower and sometimes it just never replies, is this due to the "preview" nature? is this a "known issue"? fixes?

Seems I'm not the only one with this feeling: https://www.reddit.com/r/GithubCopilot/comments/1gjdsci/copilot_chat_with_claude_35_sonnet_preview_slow/?rdt=44914