Google released gemini-3.7-flash on 2026-08-13. We graded it the same day on Antigravity CLI, on the Google AI Pro plan at effort medium. It replaces Gemini 3.6 Flash on the bench. In short: it solves more, uses less quota, and takes longer.
What the results show
- It is better on hard work: Gemini 3.7 Flash now solves three of the five hardest real-world tasks, where 3.6 Flash solved two (real-world@v3). That puts it first on the leaderboard.
- On vibe-coding@v1 nothing changed: it solved exactly the same tasks as 3.6 Flash.
- It uses less quota: a full real-world run took 9% of the Google AI Pro weekly window, down from 12.6% for 3.6 Flash.
- It is slower: the same vibe-coding tasks took about twice as long (median 3m 40s, up from 1m 54s). Real-world task time stayed about the same.
For coding on Google AI Pro this is a good trade. The same weekly quota now buys more solved hard work, and the price is waiting, not quota. If speed matters most to you, the upgrade is less clear. For everyone else it is a straight step up.
Grades are comparable within the same suite version. See the full Gemini 3.7 Flash scorecard or follow the family over time on Trends.