I will base my resolution on the largest version of Gemini 3.0 released, not counting parallel test-time compute scaling (like deep think).

Will resolve if the model comes out before 2026 but Metr doesn’t evaluate it until after.