What would you like to be added?

Currently, fastModel (configured at the top level of settings.json, or via /model --fast <id>) is only honored when the configured model id also exists under the same auth type as the main session model. If they don't match, fastModel is silently treated as unset, and all features that rely on it (session title, session recap, tool-use summary, memory relevance selection, prompt suggestions, /rename --auto, etc.) fall back to either skipping or using the main model.

I would like fastModel to be allowed to point to a model registered under any configured modelProviders.<authType> entry, not only the auth type currently selected for the main session.

Concretely, given a settings.json like:

{
  "security": { "auth": { "selectedType": "anthropic" } },
  "model": { "name": "claude-opus-4-7" },
  "fastModel": "deepseek-v4-flash",
  "modelProviders": {
    "anthropic": [{ "id": "claude-opus-4-7", ... }],
    "openai": [{ "id": "deepseek-v4-flash", ... }]
  }
}

…I would like Qwen Code to actually use deepseek-v4-flash (with its own provider's envKey / baseUrl / auth) for fast-model background tasks, while the main session continues to use the Anthropic-shaped claude-opus-4-7.

Why is this needed?

Qwen Code already supports configuring multiple auth types side by side under modelProviders — openai, anthropic, and gemini — each with its own envKey, baseUrl, and model entries. Given that this multi-provider configuration is a first-class feature, it seems reasonable for fastModel to be settable across any of those configured providers — not only the one currently selected for the main session.

When the configuration as it appears in settings.json is accepted but silently treated as unset at runtime, it's hard for the user to tell that anything is wrong: fastModel still shows up in /about and /model --fast, but the background features that depend on it quietly skip. Making fastModel honor any configured provider would make the behavior match the configuration.

Additional context

The current behavior is intentional: when fastModel references a model that isn't registered under the active auth type, it is silently treated as unset, and any background feature that depends on it (session title, session recap, tool-use summary, memory relevance selection, prompt suggestions, /rename --auto, etc.) either skips or falls back to the main model.

{
  "security": { "auth": { "selectedType": "anthropic" } },
  "model": { "name": "claude-opus-4-7" },
  "fastModel": "deepseek-v4-flash",
  "modelProviders": {
    "anthropic": [{ "id": "claude-opus-4-7", ... }],
    "openai": [{ "id": "deepseek-v4-flash", ... }]
  }
}