Google has launched Gemini 3.7 Flash, a new artificial intelligence model designed to handle software development, coding and complex AI-agent workflows. The company is positioning the latest addition to its Gemini family as a practical “workhorse” for developers and businesses that want AI systems capable of completing multi-step tasks rather than simply responding to prompts.
The release comes only about three weeks after Gemini 3.6 Flash, highlighting how quickly Google is developing its AI models as competition intensifies across coding assistants, enterprise software and autonomous AI agents. Google says the new model delivers improvements in software engineering, web development and complex knowledge work while also being more economical to run.
A major focus of Gemini 3.7 Flash is AI-assisted coding. The model has been designed to improve how AI handles software issues, debugging and code generation. Instead of merely suggesting snippets of code, it is intended to work through problems, identify errors and produce more production-ready solutions.
Google‘s testing suggests the model has made significant gains over its predecessor. On the FrontierCode 1.1 Main benchmark, Gemini 3.7 Flash scored 43.6%, compared with 34.4% for Gemini 3.6 Flash. On DeepSWE v1.1, a benchmark focused on software engineering, the newer model scored 65.3%, compared with 49% for the previous version.
For developers, the improvement is particularly relevant because AI coding tools can lose efficiency when they encounter an unfamiliar codebase or when a first attempt does not work. Gemini 3.7 Flash is designed to reason through such roadblocks, use tools and make additional attempts where necessary. The goal is to reduce the number of failed loops and the amount of manual debugging required from developers.
Google is also targeting agentic coding, where an AI system can plan a task, access tools, execute commands and check the outcome before moving to the next step. This is different from conventional chatbot use, where the user generally asks a question and receives an answer. Agentic systems are expected to carry out longer workflows with less human intervention.
The new model is therefore being integrated into Google’s developer ecosystem, including the Gemini API, Google AI Studio, Google Antigravity and Android Studio. Enterprise users can access it through Google’s Gemini Enterprise Agent Platform and Gemini Enterprise app.
Web development is another area where Google expects Gemini 3.7 Flash to make a difference. The model can generate websites and applications while paying greater attention to design requirements. Google says it is better at following visual references and design systems, allowing developers to turn mock-ups into functional interfaces with fewer rounds of prompting.
That could make the model useful beyond traditional programming. Designers, product teams and developers increasingly use generative AI to build prototypes, test ideas and create user interfaces. Better adherence to a requested design can make these workflows faster and reduce the need for developers to manually correct AI-generated output.
Cost is another important part of the launch. Google is offering Gemini 3.7 Flash at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026. The company says this is half the original price of Gemini 3.6 Flash. From January 2027, the price is scheduled to increase to $1.50 per million input tokens and $7.50 per million output tokens.
The pricing strategy matters because AI agents can consume large volumes of tokens. An agent may need several model calls to understand a request, plan its response, use external tools, examine the results and correct mistakes. Lower costs can therefore make agent-based applications more commercially viable, particularly for companies processing thousands or millions of tasks.
The development is part of Google’s broader shift from conversational AI towards agentic AI. Earlier this year, the company introduced several tools aimed at allowing AI systems to take action, including Google Antigravity and managed agents through the Gemini API. These systems are designed to combine reasoning with functions such as code execution, file management and web retrieval.
