When utility small model is not selected, Copilot fails to use BYOK provider

[LogicDaemon]

related to #322751: Idea is good, but implementation is flawed: currently if no utility model selected, Copilot just fails to respond, showing the error

No utility model is configured for 'copilot-utility-small' while the selected main model is BYOK.

Moreover, I'm using llama.cpp with the model fitted tightly to VRAM, so re-loading the model is much higher overhead than just using the currently selected one for small requests. But if I have Qwen3.6 selected, and my utility model is, for example, Gemma 4 12B, its usage will require re-loading the model. To avoid that, there should be an option "Use same model as the main selected one", and a way to select models of a different provider (currently, dropdowns for utility models only list models of the BYOK provider).

• Copilot Chat Extension Version: builtin to VS Code
• VS Code Version: 1.128.0-insider (7e7f950)
• OS Version: 10.0.26200.8655
• Feature (e.g. agent/edit/ask mode): Agent
• Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): BYOK
• Logs:

2026-07-02 11:48:59.653 [warning] [IntentDetector] Skipping participant detection context telemetry: No utility model is configured for 'copilot-utility-small' while the selected main model is BYOK.
2026-07-02 11:48:59.712 [warning] [virtual-tools] Had to drop 1 tools due to limit constraints
2026-07-02 11:48:59.713 [warning] [virtual-tools] Had to drop 1 tools due to limit constraints
2026-07-02 11:48:59.722 [error] Error: No utility model is configured for 'copilot-utility-small' while the selected main model is BYOK.
	at f_._resolveUtilityFamily (/home/user/.vscode-server-insiders/bin/7e7f9505b9ef0a7a0899794997d4b1cb608084fc/extensions/copilot/dist/extension.js:6873:16909)
2026-07-02 11:49:00.811 [info] message 0 returned. finish reason: [stop]
2026-07-02 11:49:00.813 [info] request done: requestId: [fe717424-7a8a-439a-aaca-82bc4bc7f0d0] model deployment ID: []
2026-07-02 11:49:00.814 [info] ccreq:ade8e125.copilotmd | success | gpt-4o-mini-2024-07-18 | 679ms | [title]

Steps to Reproduce:

1. Configure chatLanguageModels.json to include

	{
		"name": "llama.cpp@:8080",
		"vendor": "customendpoint",
		"apiType": "chat-completions",
		"models": [
    {
        "id": "qwen3.6-reap-moe-256k",
        "name": "Qwen3.6-28B-REAP20-A3B-256k",
        "url": "http://127.0.0.1:8080/",
        "toolCalling": true,
        "vision": true,
        "maxInputTokens": 229376,
        "maxOutputTokens": 32768
    },
    {
        "id": "gemma4-12b-256k",
        "name": "unsloth_gemma-4-12b-it-GGUF:Q6_K_XL-256k",
        "url": "http://127.0.0.1:8080/",
        "toolCalling": true,
        "vision": true,
        "maxInputTokens": 229376,
        "maxOutputTokens": 32768
    },
]

2. Open Agentic panel, select Qwen3.6-28B-REAP20-A3B-256k
3. Say "Hi", Enter

you'll get the error

No utility model is configured for 'copilot-utility-small' while the selected main model is BYOK.

4. Configure unsloth_gemma-4-12b-it-GGUF:Q6_K_XL-256k in vscode-insiders://settings/chat.utilitySmallModel
5. Retry the response

You'll see the local engine will load the second model (along with the first one if limit permits, or will unload the first model) for the utility use, then switch back to the first model, which is much slower than just generating everything with the same main model.

You may ask, why would I select qwen while gemma is configured as the utility? I frequently switch between models, including a few of my BYOK and models provided by Copilot.

The previous behavior of always using copilot models for utility worked for me mostly, but it is flawed in some parts and for example prevented me from ever using commit message generation with copilot. Also it fails when internet connectivity is missing. This in an opportunity to fix both issues!

Also perhaps, the utility model should be configurable per provider in chatLanguageModels.json, with same options as available in the user settings.json, so a cloud BYOK configuration allow setting up different chat.utilityModel and chat.utilitySmallModel, while llama.cpp config having main there (or some other non-model-name keyword) to avoid re-loading.

Either way, utility models should not be restricted to the same provider as the main selected model.

[MohammadKani]

the issue is resolved and code is merged 🙌 just need to rollout

[LogicDaemon]

محمد کنی (@MohammadKani) can you share any details, like the commit link?

[vritant24]

Hi LogicDaemon (@LogicDaemon), thanks for your feedback and easy to understand details provided :)

I do want to understand this better:

To avoid that, there should be an option "Use same model as the main selected one", and a way to select models of a different provider (currently dropdowns only show models of the BYOK provider).

Which of the following is your issue?

1. the dropdown is not showing all the models you have available
2. you would like the utility models to always be the main agent model so that when you switch the main model, you don't need to go update the utility model setting as well

[nickchomey]

Also causing my BYOK models to fail. rolled back with sudo apt install code-insiders=1.128.0-1782882825 (the latest version that isnt broken)

[vritant24]

nickchomey You can revert back to previous behavior by toggling the setting chat.useCopilotModelsForUtilityModels

Could you please provide more details on what kind of failures you're seeing?

[LogicDaemon]

To avoid that, there should be an option "Use same model as the main selected one", and a way to select models of a different provider (currently dropdowns only show models of the BYOK provider).

Which of the following is your issue?

1. the dropdown is not showing all the models you have available
2. you would like the utility models to always be the main agent model so that when you switch the main model, you don't need to go update the utility model setting as well

both.

on 1, I expected to see Copilot-provided models there (along with my BYOK models which are shown currently). For commit generation particularly, I prefer very different model from what Copilot uses by default, and for now I completely refrained from using copilot commit message generation; I still use Copilot-provided Gemini models to generate commit messages using GitLens which is usable (just try Gemini 3.1 Pro and you'll see the difference).

on 2 also yes, however "always" is exaggeration.
Currently the "utility model" dropdowns show "Default" and my BYOK models.
There could be another item: "The model selected in the chat interface" (alongside Default), which will make all utility generation with the model currently selected in the Chat window.

"Default" in that case should keep the old behavior when Copilot-provided model is selected, and behave like "Selected BYOK model" when a BYOK model is selected.

That will:

1. Keep all existing behavior unchanged for non-BYOK users
2. BYOK users will have chat fully working without selecting utility model if their main model is working (and if their main model isn't working utility model won't work either, but that won't be a surprise)
3. Both BYOK and non-BYOK users will be able to explicitly select utility model if they want

Also perhaps, the utility model should be configurable per provider in chatLanguageModels.json, with same options as available in the user settings.json, so a cloud BYOK configuration allow setting up different chat.utilityModel and chat.utilitySmallModel, while llama.cpp config having main there (or some other non-model-name keyword) to avoid re-loading.

Either way, utility models should not be restricted to the same provider as the main selected model.

[vs-code-engineering]

This feature request is now a candidate for our backlog. The community has 60 days to upvote the issue. If it receives 20 upvotes we will move it to our backlog. If not, we will close it. To learn more about how we handle feature requests, please see our documentation.

Happy Coding!

[yanke1311]

also meet same issue