Google has launched Gemini 3.7 Flash, an upgraded artificial intelligence model focused on software engineering, web development, business workflows and autonomous agents.
The model arrives three weeks after Gemini 3.6 Flash. Google describes it as its most capable Flash model for coding and agent-based tasks.
Gemini 3.7 Flash is available through the Gemini API, Google AI Studio, Google Antigravity, Android Studio and Google’s enterprise AI platforms. It is also being added to Gemini Spark for eligible Google AI Pro and Ultra subscribers.
Coding Performance Improves
Google said the model is better at debugging, resolving software issues and generating usable code on its first attempt. It is also designed to follow detailed instructions more closely and adjust its approach when it encounters a problem.
The company reported improvements across two software engineering tests.
On FrontierCode 1.1 Main, which measures production code quality, Gemini 3.7 Flash scored 43.6%. Gemini 3.6 Flash recorded 34.4% on the same benchmark.
On DeepSWE v1.1, which tests long-running software engineering work, the new model scored 65.3%, up from 49% for its predecessor.
The published benchmark results include:
- FrontierCode 1.1 Main: 43.6% versus 34.4%
- DeepSWE v1.1: 65.3% versus 49%
- WebDev Arena: Elo score of 1,588 versus 1,538
- GDP.pdf: 34% versus 22%
- AutomationBench: 30.4% versus 17%
The comparisons are against Gemini 3.6 Flash.
Stronger Web and App Development
The Google AI model can generate web pages and applications from text instructions, screenshots, images and complete design systems.
Google said Gemini 3.7 Flash produces more functional layouts and feature-complete applications with fewer prompts. The company also reported stronger adherence to reference designs when creating user interfaces.
The model reached an Elo score of 1,588 on WebDev Arena, compared with 1,538 for Gemini 3.6 Flash.
Google demonstrated the model generating a playable 3D game from a text prompt. Gemini 3.7 Flash handled the underlying workflow while Nano Banana generated characters, textures and other visual assets.
In another demonstration, the model coordinated multiple AI agents to create an interactive landing page. It also converted information from a static annual report into a web-based data presentation with charts and aggregated findings.
Agentic Workflows Receive an Upgrade
Gemini 3.7 Flash places more emphasis on multi-step planning, tool calls and task execution. It can clarify instructions when required, respond to roadblocks and continue working through a sequence of connected steps.
Google is positioning the model for software agents and enterprise automation. Potential tasks include collecting information from files, drafting messages, updating documents and carrying out processes that require several tools.
On AutomationBench, which evaluates business workflow automation, the model scored 30.4%. Gemini 3.6 Flash achieved 17%.
The model also improved its performance when handling complex professional documents. On the GDP.pdf benchmark, Gemini 3.7 Flash scored 34%, compared with 22% for the previous version.
Google said these changes are intended for knowledge-heavy work in sectors such as finance, law and biosciences. The model can process documents, extract information and use the results within a wider workflow.
Multimodal Input and One-Million-Token Context
Gemini 3.7 Flash accepts text, images, audio, video and PDF documents as inputs. Its output is currently limited to text.
The model supports an input context window of 1,048,576 tokens and a maximum output of 65,536 tokens.
Its supported capabilities include:
- Code execution
- Function calling
- File search
- Search grounding
- Structured outputs
- URL context
- Cached prompts
- Computer use in preview
- Adjustable reasoning levels
Developers can set the model’s reasoning level to low, medium or high. A minimal thinking setting is not supported.
The model has a March 2026 knowledge cutoff, although information coverage can vary by subject. Google’s model documentation also warns that Gemini 3.7 Flash may produce inaccurate information, experience occasional delays or time out during some requests.
Introductory Price Runs Until December
Google has priced Gemini 3.7 Flash at $0.75 per one million input tokens and $3.75 per one million output tokens during its introductory period.
These rates will remain available until December 31, 2026. From January 1, 2027, the listed price will increase to $1.50 per one million input tokens and $7.50 per one million output tokens.
The introductory rate is half the original per-token price of Gemini 3.6 Flash.
Gemini Spark Moves to the New Model
Gemini Spark will begin using Gemini 3.7 Flash. Spark is Google’s personal AI agent for Google AI Pro and Ultra subscribers in more than 160 countries.
The agent is designed to operate continuously under a user’s direction. It can work across supported Google Workspace services and handle tasks such as organising files, drafting emails and updating status documents.
Google said the new model improves Spark’s tool use, accuracy and output quality across workflows requiring several skills.
Gemini 3.7 Flash also includes updated safeguards addressing cyber misuse and chemical, biological, radiological and nuclear risks. Google said the model passed the required launch thresholds, but its documentation continues to list hallucinations and occasional performance issues among its known limitations.
