Google DeepMind has introduced Gemini 3.7 Flash, the latest iteration of its workhorse model family, designed to deliver enhanced performance for coding and agentic workflows. The release comes just three weeks after Gemini 3.6 Flash, reflecting rapid iteration based on developer feedback and algorithmic improvements.
The new model shows substantial gains in software engineering tasks, including debugging and issue resolution. On the FrontierCode 1.1 Main benchmark, it scores 43.6% compared to 34.4% for 3.6 Flash, and on DeepSWE v1.1 it reaches 65.3% versus 49.0%. For web development, it achieves an Elo of 1588 on WebDev Arena, up from 1538, and generates more functional layouts with fewer prompts.
In knowledge-intensive domains such as finance, law, and biosciences, Gemini 3.7 Flash demonstrates improved reasoning and accuracy. It scores 34.0% on the GDP.pdf benchmark versus 22.0% for 3.6 Flash, and 30.4% on AutomationBench compared to 17.0%, indicating better performance in real-world business workflows.
The model is priced at $0.75 per million input tokens and $3.75 per million output tokens, an introductory rate that is half the cost of 3.6 Flash. This pricing, combined with enhanced performance, aims to enable developers to scale production-ready agents cost-effectively.
Gemini 3.7 Flash also powers Gemini Spark, the 24/7 personal agent available to Google AI Pro and Ultra subscribers in over 160 countries. The update improves tool use for Google Workspace apps, making Spark more efficient for knowledge work. Safety safeguards have been updated for CBRN and cyber domains, in line with Google's responsible AI approach.