Half the cost, maximum performance: Google launches Gemini 3.7 Flash
Google is launching an innovative working model designed for advanced coding and autonomous agents. The upgrade arrived just three weeks after the previous version and includes a leap forward in bug fixing and complex document processing. Alongside performance improvements, the company is cutting rates by 50% to facilitate AI integration in organizations.
Just three weeks. That is the time it took for tech giant Google to land its next blow in the artificial intelligence race and launch Gemini 3.7 Flash. The AI market has become accustomed to a fast update pace, but this time it is not just another minor fix. Google is presenting an algorithmic leap focused on coding skills, automation of complex processes, and the operation of autonomous agents.
Behind the rapid launch is a combination of aggressive feedback from the developer community and breakthroughs in multi-step planning. The result? A core working model that not only achieves higher results in international performance benchmarks but also cuts the price by 50% compared to the 3.6 Flash version at its launch.
Ruthless coding engine
The main upgrade in 3.7 Flash is felt directly in software development environments. The new model demonstrates a much higher analytical capability in fixing bugs, dealing with obstacles during code writing, and understanding complex instructions on the first attempt.
In website and user interface (UI) development, the model is capable of taking a screenshot, a sketch, or an entire design system and translating them into operational and complete code with fewer commands.
In the DeepSWE v1.1 test for detecting and resolving software faults, the model recorded a jump to 65.3% compared to 49.0% in the previous version. In the FrontierCode 1.1 Main index, the model climbed to 43.6% (compared to 34.4%). In the WebDev Arena index from Arena.ai, the new version recorded an Elo score of 1588 points.
"The model thinks more rigorously, investing significant effort in multi-step planning and tool usage," said Tulsee Doshi, Senior Director of Product Management at Gemini. "Precise execution means less manual intervention by engineers and fewer retries throughout the development process."
Beyond code
The autonomous capabilities of Gemini 3.7 Flash are not limited to the software world. Google has aimed the model at handling knowledge-intensive data pipelines in the fields of law, finance, and life sciences. By increasing accuracy in processing complex documents, the model is capable of turning static financial reports in PDF format into interactive data stories that include dynamic graphs and processed insights.
Thus, the GDP.pdf test (understanding complex documents) jumped to 34.0% compared to 22.0% in the 3.6 Flash version. In the AutomationBench test (business automation), one can see a jump to 30.4% in the ability to complete end-to-end processes in the real world.
Price wars
Alongside the technological upgrade, Google is landing an economic blow on its competitors. Until the end of the year, the model is offered at an aggressive introductory price. Cost reduction is a critical element for organizations seeking to run AI agents that operate in the background 24/7. Simultaneously, Google updates that its personal AI agent, Gemini Spark (available to Google AI Pro and Ultra subscribers), will be powered from today by the new model, which will dramatically improve synchronization and automation with Google Workspace applications.
Google notes that alongside the performance upgrade, protection mechanisms have been integrated to prevent misuse of the model in the fields of chemical, biological, nuclear, and offensive cyber weapons, while maintaining protected analytical capabilities for security companies.
The model is available starting today for developers via the API in Google AI Studio, Android Studio, and the Google Antigravity environment, as well as for enterprise customers in Gemini Enterprise.
