Introducing Gemini 3.7 Flash

Google is introducing Gemini 3.7 Flash as its most intelligent workhorse model yet, designed specifically for coding and autonomous agents. Released just three weeks after the previous version, the new model brings substantial performance improvements to software engineering, knowledge work, and web development workflows.

In coding tasks, the model achieves higher first-pass code accuracy, better debugging, and improved production-ready code generation, showing notable gains on benchmarks like FrontierCode and DeepSWE. For web development, it creates functional layouts and feature-complete applications using fewer prompts, outperforming the prior version on WebDev Arena. It also delivers stronger reasoning in knowledge-dense fields such as law, finance, and biosciences, scoring significantly higher on complex document processing and real-world business workflow benchmarks.

The developer experience is enhanced through better instruction following, multi-step planning, and disciplined execution that reduces the need for manual oversight. Through the end of the year, the model is available at an introductory price of half the cost of the previous version, charging $0.75 per million input tokens and $3.75 per million output tokens.

Gemini Spark, the 24/7 personal agent available to Google AI Pro and Ultra subscribers, now uses the new model to handle knowledge work and Google Workspace tasks more efficiently. Additionally, the model ships with updated Frontier Safety safeguards against misuse in chemical, biological, radiological, nuclear, and cyber offense domains. It is accessible immediately to developers via Google AI Studio, Android Studio, and Google Antigravity, as well as to enterprises and individual subscribers.