Gemini 3.7 Flash lands in the Gemini app, at half the price

Google shipped Gemini 3.7 Flash on 13 August and it's already replacing 3.6 Flash in the Gemini app's model picker across Android, iOS, web and macOS. It hit Gemini Spark first (AI Pro and Ultra subscribers, 160+ countries), then rolled into regular chats. If your picker still says 3.6, it's a staged rollout — wait.

The numbers explain the hurry. On the Artificial Analysis Intelligence Index it scores 56 — a 4-point jump over 3.6 Flash, one point behind GPT-5.6 Terra and Muse Spark 1.2, but at an average time per task of 1.7, roughly 40% faster than Terra at max reasoning. Coding is where the gap really opens: FrontierCode 1.1 goes 34.4% → 43.6%, and DeepSWE v1.1 jumps 49.0% → 65.3%. On AutomationBench, the enterprise-workflow benchmark, it nearly doubles to 30.4% against Claude Sonnet 5's 10.7%.

The aggressive part is the price. Google halved token cost through 31 December 2026 and pushed the model into the Gemini API, the Enterprise Agent Platform, and GitHub Copilot on day one. That's a land-grab, not a launch — three weeks after 3.6 Flash, Google is betting developers switch on price-per-point before rivals reprice.

Why it matters: the frontier is now a workhorse tier, and Google is willing to lose margin to own it.