Feature request: GPT-4o Copilot code completion model

[helto4real]

With the new code completion model just released it would be nice to be able to select that model as completion model.

https://github.blog/changelog/2025-02-18-new-gpt-4o-copilot-code-completion-model-now-available-in-public-preview-for-copilot-in-vs-code/

Big thanks to all contributors for this plugin!

[zbirenbaum]

This looks like a really nice feature, definitely want to include it. I'll take a look at how to integrate it. I'm a bit busy and may take a few days so in the mean time anyone else is also welcome to take a stab at it if they are inclined since

[helto4real]

Sounds awesome, wish I had the skill I would indeed ;)

[Nikola-Milovic]

I stumbled upon this exact question and it seems it was solved less than an hour ago in copilot.vim @zbirenbaum

https://github.com/orgs/community/discussions/151719#discussioncomment-12281578

[trixnz]

I managed to get it working using the following Lazy config:

return {
  {
    "zbirenbaum/copilot.lua",
    branch = "create-pull-request/update-copilot-dist",

    config = function(_, opts)
      local new_opts = vim.tbl_extend("force", opts, {
        server_opts_overrides = {
          cmd = {
            "node",
            vim.api.nvim_get_runtime_file("copilot/dist/language-server.js", false)[1],
            "--stdio",
          },
          init_options = {
            copilotIntegrationId = "vscode-chat",
          },
        },
      })

      require("copilot").setup(new_opts)

      local util = require("copilot.util")
      local orig_get_editor_configuration = util.get_editor_configuration

      ---@diagnostic disable-next-line: duplicate-set-field
      util.get_editor_configuration = function()
        local config = orig_get_editor_configuration()

        return vim.tbl_extend("force", config, {
          github = {
            copilot = {
              selectedCompletionModel = "gpt-4o-copilot",
            },
          },
        })
      end

      return new_opts
    end,
  },
}

Integrating #368 would require a few minor changes:

• Update the node invocation from index.js to language-server.js with --stdio.
• Pass copilotIntegrationId = "vscode-chat", in the LSP init_options. Although tpope suggests this may not be necessary in the future.
• Updating the editor configuration to include:

   github = {
     copilot = {
       selectedCompletionModel = "gpt-4o-copilot",
     },
   },

I suppose with the release of the copilot language server we could probably target that instead of pulling from the copilot.nvim plugin at some point. I assume it would have a faster release cadence.

[9mm]

Does anyone happen to know the what existing model copilot has always used?

[bagaze]

Does anyone happen to know the what existing model copilot has always used?

GPT 3.5 Turbo. See https://docs.github.com/en/copilot/using-github-copilot/ai-models/changing-the-ai-model-for-copilot-code-completion?tool=vscode#overview

[AntoineGS]

I managed to get it working using the following Lazy config:
...

@trixnz, how are you checking which model Copilot is using?
I can't seem to figure out how to get confirmation that I am, even the Info logs from the LSP do not seem to log it. :(

Edit: I figured that the lack of error meant it was using the right model, as using an incorrect model name does give an error.

[9mm]

I get the error "editorInfo and editorPluginInfo will soon be required in initializationOptions. this will replace setEditorInfo"

Do you happen to know how i fix this? ive searched and dont see much about it on google or in this repo, but it seems to definitely stem from copilot somehow

[AntoineGS]

Ah that's just a warning which seems to be always output even when doing what it asks 🤣 I will see what I can do and maybe scale back the logging.