Unify reasoning selection for providers with separate model variants

[NikitaMGrimm]

Note

🤖 GPT-6.1 Sol writing on behalf of NikitaMGrimm

Problem

Some providers encode reasoning effort in separate model IDs. Antigravity, for example, lists Gemini 3.8 Flash (High), Gemini 3.8 Flash (Medium), and Gemini 3.8 Flash (Low) as three models.

To change only the reasoning level, users have to choose another model entry. This makes the catalog longer and gives reasoning selection a different workflow from providers that expose it as a model option.

Current mobile picker

The screenshot shows the three Antigravity entries. It was supplied by the user; the app version was not recorded.

Proposed behavior

Show one Gemini 3.8 Flash entry with a Reasoning selector offering the variants available from that provider: Low, Medium, or High. Choosing a level would select the corresponding native model ID.

Use the same interaction across web, desktop, and mobile. Preserve existing saved selections and favorites, and only group variants when their relationship is known. Models without a known effort mapping should keep their existing entries.

Current behavior and direction

Current upstream code explicitly preserves Antigravity's native model IDs, including thinking levels, and exposes no separate model options. This proposal changes how those choices are presented while retaining the provider's IDs.

Related: #9478 and #9511 concern missing or stale Antigravity model choices, rather than grouping reasoning variants.

Would maintainers support this direction, starting with Antigravity's known variants?