Google launched Gemini 3.7 Flash just three weeks after its last release and continues to accelerate its pace of model updates. new version, according to 9 to 5 googlewith improvements in programming, web development, and handling complex problems.
Image source: Google
Google says Gemini 3.7 Flash is “the result of developer feedback and algorithm innovation.” The most significant changes were observed when working with code, where the model is now better able to find and correct errors. In the DeepSWE v1.1 test, its results improved from 49.0% to 65.3% in Gemini 3.6 Flash, and from 34.4% to 43.6% in FrontierCode 1.1 Main.
These changes also affect web development. This model enables the creation of more functional interfaces and applications, with more functionality out of the box, but with fewer requests. In WebDev Arena, the score of Gemini 3.6 Flash increased from 1538 points in the previous version to 1588 points. Gemini 3.7 Flash also better retains the characteristics of the original design when creating interfaces, whether the reference is screenshots, images or mature design systems.
Google singled out improvements for tasks involving large amounts of professional information, including finance, law and life sciences. In the GDP.pdf test, which tests the model’s ability to handle complex files, the Gemini 3.7 Flash scored 34.0%, while the 3.6 Flash scored 22.0%. In AutomationBench, a multimodal benchmark that evaluates real-world workflow execution, results also increased—from 17.0% to 30.4%.
One of the most noticeable changes is the pricing policy. Access to Gemini 3.7 Flash will be provided to developers at $0.75 per 1 million input tokens and $3.75 per 1 million output tokens by the end of 2026. According to Google, this is half the original price of Gemini 3.6 Flash. Starting January 1, 2027, the price will increase to $1.50 per million input tokens and to $7.50 per million output tokens.
Gemini 3.7 Flash has begun rolling out to Spark’s personal AI agents, but requires an AI Pro or Ultra subscription to access. The model is also available in Google Antigravity, AI Studio, Android Studio, Gemini Enterprise Agent Platform, and Gemini Enterprise applications. Google expects these improvements will help AI agents handle work tasks and interact with Google Workspace apps more efficiently.
If you find an error, select it with your mouse and press CTRL+ENTER.
