Gemini 3.7 Flash reaches AI Mode, a day after its release

Google’s AI Mode picked up a new engine on 14 August 2026: Gemini 3.7 Flash, selectable from the model menu a single day after Google shipped the model itself. Robby Stein, Google’s VP of Product for Search, announced the rollout directly, describing the model as “better at following instructions and understanding your intent, so you get even more helpful responses.”

What actually changed#

Gemini 3.7 Flash replaces the prior “workhorse” model behind AI Mode’s default reasoning for subscribers who opt in. It carries a large input context window and a knowledge cutoff of March 2026, which matters for how much of a long page or conversation the model can hold in view when it answers. None of Google’s public description touches ranking, indexing, or how sources get chosen; the change is confined to how the model reasons over and phrases the answer once retrieval has already happened.

Who can use it, and how#

The model is gated behind a Google AI Pro or Ultra subscription and is live globally in English. Reaching it takes a manual step: opening the plus icon in AI Mode’s “Ask anything” bar and picking Gemini 3.7 Flash from the model menu rather than the default option. Free-tier AI Mode users do not see the option at all.

Why we care#

A model swap inside AI Mode is invisible in every ranking-tracking tool a site runs, because it changes phrasing and reasoning quality, not which pages get pulled in. It is worth watching for a narrower reason: Google has now shipped two model generations into AI Mode inside a single year, which means the answer layer above your citations is on a materially faster upgrade cycle than the ranking layer underneath it.