Google introduced Gemini 3.7 Flash, a model designed for programming, AI agents, and complex workflows. It arrives just three weeks after Gemini 3.6 Flash, but with important improvements in code generation, web development, document analysis, and business automation.
The most striking detail is that Google is also lowering the introductory price: until the end of the year, Gemini 3.7 Flash will cost $0.75 per million input tokens and $3.75 per million output tokens. What does this mean in practice? Teams can run more capable agents without every task leading to a significant increase in costs.
More precision for coding and problem-solving
Gemini 3.7 Flash improves on version 3.6 in tasks such as debugging, issue resolution, and generating production-ready code. Google says the model delivers greater accuracy on the first attempt, which is especially valuable when a tool needs to act with limited human supervision.
In the evaluations cited by the company, it achieved the following results:
- FrontierCode 1.1 Main: 43.6% compared with 34.4% for Gemini 3.6 Flash.
- DeepSWE v1.1: 65.3% compared with 49.0% for Gemini 3.6 Flash.
These numbers do not mean the model can automatically replace an engineering team. They do suggest that it can reduce the number of fixes, retries, and reviews needed for well-defined development tasks.
Web development with fewer instructions
The new version also aims to create more functional interfaces and feature-rich applications using fewer prompts. For interface generation, Google highlights a stronger ability to follow a reference design, whether it is a screenshot, an image, or a complete design system.
In WebDev Arena, an evaluation by Arena.ai for comparing models in web development, Gemini 3.7 Flash reached an Elo score of 1588, above Gemini 3.6 Flash's 1538 points.
For a developer or entrepreneur, the difference can be felt in concrete tasks: turning a sketch into a navigable interface, maintaining a brand's colors and spacing, or generating a functional first version of an admin dashboard without having to describe every detail several times.
Better performance in documents and automation
Gemini 3.7 Flash also improves in knowledge-dense areas such as finance, law, and biosciences. On the GDP.pdf benchmark, designed to evaluate the comprehension of complex documents, it scored 34.0%, compared with 22.0% for Gemini 3.6 Flash.
In AutomationBench, an evaluation focused on real-world business workflows, it achieved 30.4%, compared with 17.0% for the previous version. The figure suggests a stronger ability to complete multi-step processes, although any use in regulated fields still requires professional validation.
A more capable model does not eliminate the need for review. The advantage is that it can move further before human intervention is needed.
Agents with better planning and less supervision
Google describes Gemini 3.7 Flash as a model that adapts better when it encounters obstacles, asks for clarification when the intent is unclear, and follows instructions more faithfully.
It also devotes more effort to multi-step planning and tool use. In an AI agent, this is essential: it is not enough to answer a question; the agent must also decide what action to take, in what order, and how to verify the result.
For example, an agent could gather information from several files, draft an email, and update a status document. If it misinterprets an instruction at any of those steps, the final result loses value. Gemini 3.7 Flash promises to reduce these errors and the number of retries required.
Gemini Spark gets the update
Gemini Spark will begin using Gemini 3.7 Flash today. According to the company, this personal Google agent is available to Google AI Pro and Ultra subscribers in more than 160 countries.
Spark can work with Google Workspace apps and help turn ideas into actions. The examples mentioned include consolidating files, drafting emails, and updating tracking documents.
With the new model, Google expects to improve the accuracy and quality of results in tasks that combine several skills. The difference will be especially relevant when users are not looking for an isolated answer, but want to complete a process from start to finish.
Safety for sensitive uses
Gemini 3.7 Flash includes updates to its safety measures against misuse related to chemical, biological, radiological, and nuclear risks, as well as cyberattacks.
Google says these protections are intended to limit harmful uses without blocking beneficial applications. Additional technical details are available in the model's model card, which documents its capabilities, risks, and safety evaluations.
Where to try Gemini 3.7 Flash
- Developers: can explore agent-focused workflows through Google Antigravity or get started with the Gemini API in Google AI Studio and Android Studio.
- Businesses: can access the model through Gemini Enterprise Agent Platform and the Gemini Enterprise app.
- Individual users: can use it through Spark in the Gemini app, as long as they have a Google AI Pro or Ultra plan in a supported country.
The arrival of Gemini 3.7 Flash shows where the industry is competing: models that do not just generate text or code, but also plan, use tools, and complete tasks with less supervision. The key will not only be the model's power, but how much time and money it can save when responsibly integrated into a real workflow.
Original source
https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash