New models not showing up in Emacs

[eamonnsullivan]

My company has recently expanded the models we have access to, opening up a few premium ones (with XX premium requests a month). Although the new models are selectable in the online version of copilot and my VS Code-using colleagues can see and select them (including an option called Auto), I still see only the one GPT-4.1 Copilot model when I try M-x copilot-select-completion-model. Is there a cache or a setting somewhere that I need to modify? I've tried logging out and back in again a few times. It's only been about 24 hours since the new models were enabled. I'm using the latest version (on straight.el) and the master branch of emacs-plus on Mac. I even tried asking copilot, which said I don't see those models because Emacs isn't supported... I don't think that's really the case, since the VS Code colleagues are using the same language server backend.

[eamonnsullivan]

I've tried setting other models manually (e.g., (setopt copilot-completion-model ...)), but I see something like this in the language server logs.

[default] User selected model auto is not in the list of generic models: gpt-41-copilot, falling back to default model.

(I've tried others besides auto, but auto is highly recommended because it saves requests.)

[eamonnsullivan]

Here's a more complete example log from the language server:

[lsp] GitHub Copilot Language Server 1.457.1 initialized
[certificates] Removed 20 expired certificates
[default] Policy watcher started for GitHub Copilot Plugin
[GithubAvailableEmbeddingTypes] Got embeddings. Primary: metis-1024-I16-Binary. Deprecated: 
[AutoModelService] Auto model cache cleared due to token update
[Semantic Search] Semantic search is disabled { reason: 'watchedFiles capability is not available' }
[default] User selected model gtp-4o-copilot is not in the list of generic models: gpt-41-copilot, falling back to default model.
[default] User selected model gtp-4o-copilot is not in the list of generic models: gpt-41-copilot, falling back to default model.

[bbatsov]

This is a server-side limitation, not something copilot.el can control. The language server's copilot/models endpoint classifies models by scope — and currently only gpt-41-copilot has the "completion" scope. Models like Claude, Gemini, GPT-4o, etc. are all chat/edit-only on the server side and can't be used for inline completions.

VS Code uses the exact same endpoint and protocol under the hood. The broader model selection you see there is for the chat UI, not for inline completions. The "auto" option is also a chat-specific feature.

Your LS log confirms this — when you set gpt-4o-copilot manually, the server rejects it because it's not in the completion models list:

User selected model gtp-4o-copilot is not in the list of generic models: gpt-41-copilot, falling back to default model.

Nothing we can do on the client side here. If/when GitHub adds more completion-scoped models, they'll show up automatically in copilot-select-completion-model.

[eamonnsullivan]

Thank you! I did not know this.