Introducing the Next-Gen Code Completions Model for GitHub Copilot 🚀

[Akash1134]

We are thrilled to announce the launch of our new code completion model, GPT-4o Copilot, now available for VS Code and JetBrains users! 🎉

This new model, based on GPT-4o mini, has additional training on over 275,000 high-quality public repositories in over 30 popular programming languages. Users can expect enhanced quality completions with improved suggestion accuracy and length, along with better performance.

⭐ Getting Started with GPT-4o Copilot in VS Code

The model is available today to all VS Code users of GitHub Copilot. To get started, open the Copilot menu in the VS Code title bar, select Configure Code Completions... > Change Completions Model.... Alternatively, open the Command Palette and select GitHub Copilot: Change Completions Model....

⭐ Getting Started with GPT-4o Copilot in JetBrains IDEs

To get started, ensure that you are on the latest stable release of the extension (v1.5.35 or above). Click the Copilot icon in the JetBrains IDE of your choice and select Edit model for Completion. This will open the GitHub Copilot settings, where you can switch between models.

⭐ GPT-4o Copilot Access Notes

• Copilot Business or Enterprise Users: Your administrator must enable this model for your organization by opting into Editor Preview Features in the Copilot policy settings on GitHub.
• Copilot Free Users: Using this model will count toward your 2,000 free monthly completions. Learn more.

Please share your feedback as you try out the new model. It will help us improve the experience for all Copilot users. For questions, please comment below.

Happy coding with GPT-4o Copilot! 🚀

[garymjr]

Will this completion model be coming to copilot.vim as well?

[garymjr]

Claude 3.7 isn't a completion model - it's a chat model. If you're using something like CopilotChat or CodeCompanion, you can already choose it.

[tpope]

This is now working on Copilot.vim 1.44.0 without any need for a workaround. Posting one more time for posterity:

let g:copilot_settings = #{selectedCompletionModel: 'gpt-4o-copilot'}

For init.lua:

vim.g.copilot_settings = { selectedCompletionModel = 'gpt-4o-copilot' }

[hhaidrr]

@garymjr @tpope Thank you both for the clarification. Much appreciated!

[wazzup-bro]

I'm not sure what the official story will be but with the latest version of Copilot.vim you should be able to access it. Add this to your vimrc:

let g:copilot_settings = #{selectedCompletionModel: 'gpt-4o-copilot'}

Or init.lua:

vim.g.copilot_settings = { selectedCompletionModel = 'gpt-4o-copilot' }

Edit: This needs Copilot.vim 1.44.0 to work.

@tpope
What are the available model names that can be used in selectedCompletionModel?
Thank you.

[tpope]

Currently the only supported value is gpt-41-copilot, which is already the default. Previews of code completion models are generally infrequent and short-lived.

[hmorneau]

Not bad, but not great. When can we have something at the level of Cursor tab feature? Maybe you should give it a try and you will see what I mean. If for example I change var="1.2.3" to var="1.2.4" then go in a next file where var="1.2.3", Cursor will just suggest me to TAB straight to it, then an other TAB to make the change and it will become var="1.2.4". Copilot, doesn't do that, not even the latest version.

[bpto]

What about visual studio 2022 ?

[MrSimonC]

New announcement from Microsoft: https://devblogs.microsoft.com/visualstudio/gpt-4o-copilot-code-completion-model-available-now-in-visual-studio-public-preview/

[smdev24x]

Has anything changed here? I cannot choose model anymore? (pro subscriber, windows 11, insiders vsc)

[kasavin]

Has anything changed here? I cannot choose model anymore? (pro subscriber, windows 11, insiders vsc)

Hi - the model is now generally available and providing all code completions. The 3.5T model is retired, so the model picker is no longer visible.

[Saviru]

The 4o model is good but personally I recommend using Claude 3.7 thinking model. It's better than other model. Also it over powers 4o when it comes to debugging

[Jimgitsit]

IMHO, GPT-4o has significantly reduced the quality of completions. Why can we not select the previous model (or any other model for that matter) for completions? 4o is a downgrade for completions.

[kasavin]

Hi @Jimgitsit, thank you for your feedback. Can you share some examples where the model is not producing the completions you expect, either here or in the feedback tool in-product?

We don't have multiple completions models on the roadmap at the moment.

[Jimgitsit]

I'll try to capture some specific examples. But it's mostly abstract in that it's suggestions are just "dummer" than they used to be. It will spit out nonsense code when it seems obvious what the correct suggestion should be based on nearby code. It used to be much more context aware than it is now. I've been using copilot every day for the last couple of years and It was a very noticeable change that started with this GPT-4o update. Please keep in mind, I'm only talking about completions, which is all I really use or want.

Here's one that it consistently gets wrong every time:

This is an error. I have to delete the extra parenthesis every time. It did not do this before 4o.

I had one yesterday where I was using the string 'ios' in a couple places and then it suggested some code that used 'iOS' instead. Wasted a good 10-15 minutes tracking that bug down.

[Jimgitsit]

I don't have the feedback option like in your screenshot btw. Is that VS Code? I'm using Webstorm.

[Jimgitsit]

The suggestion doesn't make sense given the first part of the statement, and it wants to add an extra curly bracket. It didn't use to do stuff like this, I swear.