Google has introduced the new Gemini 3.7 Flash model; a model that, according to the company, is the most powerful Flash-series version in programming and AI agents. This release comes only three weeks after the introduction of Gemini 3.6 Flash and indicates that competition among major AI companies is increasingly shifting toward developing models that can carry out complex, multi-step tasks autonomously, not merely provide text responses.
Leap in programming
Software development is a main focus of this model; Gemini 3.7 Flash performs better in debugging, identifying issues, and producing code closer to operational environment standards. In the FrontierCode 1.1 Main test it scored 43.6 percent versus 34.4 percent for the previous version, and in the DeepSWE v1.1 test (evaluating long-term programming tasks) it reached 65.3 percent, while Gemini 3.6 Flash had only scored 49 percent. Google emphasizes this model also performs better in managing interprocess development issues and reduces the need for continuous developer intervention.
Improvements in web and app building
With fewer prompts, Gemini 3.7 Flash can produce more functional layouts and more complex web applications. In the WebDev Arena test it achieved a 1588 Elo score versus 1538 for the previous version. This model can also reconstruct reference designs from screenshots, images, or a full design system.
From assistant to autonomous agent
The main focus of this model is on "AI agents"; systems that can plan steps to reach a goal, use various tools, and complete tasks with less human oversight. Google says the new model has improved at identifying obstacles, understanding the user's goal, following instructions, and planning multi-step tasks. These improvements have also reached cognitive and business domains; in the GDP.pdf complex document processing test the score rose from 22 to 34 percent and in AutomationBench (performing real workplace tasks) it increased from 17 to 30.4 percent.
Price half that of the previous version
Google has set the introductory price for Gemini 3.7 Flash through the end of the calendar year at $0.75 per one million input tokens and $3.75 per one million output tokens, which is half the introductory price of version 3.6. Starting January 1, 2027 these rates will increase to $1.50 and $7.50 respectively; a matter that has economic significance for companies that deploy AI agents at large scale.
Presence in Google products and competition with OpenAI and Anthropic
This model is now available in the Gemini Spark service for Google AI Pro and Ultra subscribers in more than 160 countries, as well as in Google AI Studio, the Gemini API, Android Studio, and Gemini Enterprise. The release of this model comes amid Google's increasing competition with OpenAI and Anthropic, although the release date of the flagship Gemini 3.5 Pro model is still unknown. Google also reported updates to the model's safety mechanisms against cyber misuse and chemical, biological, radiological, and nuclear threats.