Google has released Gemini 3.7 Flash, three weeks after 3.6 Flash. It calls the model its "most intelligent workhorse" for coding and agents.
The gains are largest on software engineering work. On DeepSWE v1.1, which tests long-horizon engineering tasks, the score rises from 49.0% to 65.3%. On FrontierCode 1.1 Main it moves from 34.4% to 43.6%. Web development is up too: 1588 Elo on WebDev Arena, against 1538 for the older model.
Two other numbers stand out. Document comprehension on the GDP.pdf benchmark jumps from 22.0% to 34.0%. Enterprise workflow automation nearly doubles, from 17.0% to 30.4%.
What it costs
Until 31 December 2026 the introductory rate is $0.75 per million input tokens and $3.75 per million output tokens. After that it doubles, to $1.50 and $7.50. Google says this is half the price of 3.6 Flash.
Where you can use it
Developers get it in Google AI Studio and Android Studio. Enterprises get it through the Gemini Enterprise Agent Platform and the Gemini Enterprise app. Individual subscribers on Google AI Pro and Ultra get it in Gemini Spark, across more than 160 countries.
The short gap between releases is the thing to watch. If you benchmarked 3.6 Flash for a project, those numbers are already out of date.
Source: Google DeepMind
