Google has introduced Gemini 3.7 Flash, describing it as its most capable Flash model yet for coding, autonomous agents and complex knowledge work.
The new model arrives just three weeks after Gemini 3.6 Flash and brings substantial improvements across software engineering, web development, document reasoning and business automation.
Google is also launching Gemini 3.7 Flash at an introductory price of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens through the end of the year, roughly half the original Gemini 3.6 Flash pricing.
Gemini 3.7 Flash Delivers Major Coding Improvements
Google said Gemini 3.7 Flash shows significant gains over its predecessor in debugging, issue resolution and first-pass code accuracy.
On FrontierCode 1.1 Main, the model scored 43.6%, compared with 34.4% for Gemini 3.6 Flash.
The improvement was even larger on DeepSWE v1.1, where Gemini 3.7 Flash reached 65.3%, up from 49.0%.
Google said the model is also better at producing production-ready code with fewer retries, while following developer instructions more consistently.
The company said Gemini 3.7 Flash is more capable of adapting when it encounters roadblocks, clarifying user intent when necessary and executing multi-step plans involving tools with less manual supervision.
Google Improves Web Development and UI Generation
Gemini 3.7 Flash also brings notable improvements to web development workflows.
Google said the model can generate more functional layouts and feature-complete applications with fewer prompts.
For user interface generation, the model is designed to adhere more closely to reference material, including screenshots, images and complete design systems.
On Arena.ai’s WebDev Arena, Gemini 3.7 Flash achieved an Elo score of 1,588, compared with 1,538 for Gemini 3.6 Flash.
The improvement could make the model more attractive for developers using generative AI to rapidly prototype or build production-facing websites and applications.
Finance, Law and Biosciences Reasoning Also Improve
Google is also positioning Gemini 3.7 Flash as a stronger model for knowledge-intensive professional workloads.
The company said it delivers higher reasoning accuracy in fields including finance, law and biosciences.
On GDP.pdf, a benchmark designed to test a model’s ability to process complex documents, Gemini 3.7 Flash scored 34.0%, up from 22.0% for Gemini 3.6 Flash.
The model also posted a significant gain on AutomationBench, reaching 30.4% compared with 17.0% for its predecessor.
Google said the AutomationBench improvement reflects stronger performance when completing real-world business processes involving multiple steps and tools.
Gemini 3.7 Flash Priced at $0.75 Per Million Input Tokens
Cost is another major part of Google’s positioning for the new model.
Gemini 3.7 Flash will be available through the end of 2026 at an introductory price of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens.
Google said the combination of lower pricing and improved intelligence is intended to make it easier for developers to deploy production-grade AI agents at scale.
The company is increasingly competing on both model performance and inference economics as businesses evaluate the cost of running persistent AI agents and high-volume applications.
Gemini Spark Upgrades to Gemini 3.7 Flash
Google is also immediately bringing the new model to Gemini Spark.
Gemini Spark, available to Google AI Pro and Ultra subscribers in more than 160 countries, will begin using Gemini 3.7 Flash starting today.
Google launched Spark as a personal AI agent capable of operating continuously and taking actions on a user’s behalf while remaining under their direction.
The Gemini 3.7 Flash upgrade is designed to improve Spark’s performance in knowledge work and strengthen its ability to use Google Workspace tools across complex, multi-skill workflows.
The release underscores Google’s rapid model-development cadence, with Gemini 3.7 Flash arriving only weeks after its predecessor as the company continues pushing deeper into coding agents and autonomous workplace AI.
