Google has taken a new step in its “Flash” line of artificial intelligence models, introducing Gemini 3.7 Flash, the most advanced and powerful version for coding and agents. Released just three weeks after its predecessor, Gemini 3.6 Flash, this update is the direct result of feedback from the developer community and continued algorithmic innovation. Gemini 3.7 Flash brings notable improvements in software engineering, thought-intensive tasks, and web development workflows, and comes with a promotional launch price that is half the original 3.6 Flash cost per million tokens.
Improved intelligence for complex tasks
The Flash 3.7 model demonstrated notable improvements compared to Flash 3.6 in coding activities, including debugging and troubleshooting. It can generate code with greater accuracy on the first try and exhibits superior performance in creating code ready for production environments. This is evidenced by the results in the FrontierCode 1.1 Main (43.6% versus 34.4%) and DeepSWE v1.1 (65.3% versus 49.0%) benchmarks, showing their effectiveness in real development scenarios.
On the same topic: Google’s Gemini Spark AI service reaches more users with paid plans
In the web development sector, Gemini 3.7 Flash makes it possible to create more practical layouts and complete applications with fewer steps. When it comes to generating user interfaces, the model achieves high conformity and design equivalence, starting from various references, such as screenshots, images or already established design systems. Its performance is superior to that of Flash 3.6 in Arena.ai’s WebDev Arena, where it obtained an Elo score of 1588 against 1538.
For knowledge-intensive industries such as finance, law and biosciences, Flash 3.7 delivers improved reasoning and accuracy. The model performed significantly better than Flash 3.6 in the GDP.pdf benchmark (34.0% versus 22.0%), which assesses the ability to process complex documents. It has also been shown to outperform Flash 3.6 in AutomationBench, proving its efficiency in completing real business workflows (30.4% vs. 17.0%).
Practical model use cases
The model demonstrates the ability to go from a simple text message to creating a fully playable 3D game. To achieve this, Gemini 3.7 Flash is combined with Nano Banana, enabling dynamic, real-time generation of characters, items and textures.
Another application is the generation of interactive and visibly impactful landing pages, carried out in a single step. Gemini 3.7 Flash orchestrates the subagents, and Gemini Omni is employed to develop fluid and interactive parallax components.
In robotics, a robotic model can be trained with Gemini 3.7 Flash. It uses multimodal understanding in a three-agent graph system, which significantly accelerates robot learning.
More on this story: Expansion of Gemini Spark virtual assistant requires premium Google subscriptions
There is also the ability to transform a static PDF into an interactive data narrative. Complex annual reports are converted into captivating web experiences that include dynamic charts and compiled insights.
Improved development experience and pricing
Gemini 3.7 Flash provides a significantly better development experience compared to its predecessor. The model demonstrates a greater ability to adapt to challenges, clarifying intentions when necessary, and following instructions more accurately. Your way of “thinking” is more dedicated, with more in-depth planning in multiple steps and optimized use of tools. This more disciplined execution results in less need for manual oversight and reduced repeated attempts at engineering processes.
Full coverage: Latest News (EN)
Flash 3.7 is expected to be available before the end of the year, with an attractive launch price of $0.75 per million tokens in and $3.75 per million tokens out. This pricing, combined with the model’s superior performance, makes it easier for developers and companies to implement AI agents on a large scale for production, in a financially advantageous way.
Initial customer feedback already highlights the performance and accuracy of Flash 3.7, indicating considerably better results than Flash 3.6, at a reduced cost.
The impact of Gemini 3.7 Flash on Gemini Spark
Learn more: Google expands access to Gemini Spark, but artificial intelligence service will be paid
Gemini Spark, a tool accessible to Google AI Pro and Ultra subscribers in more than 160 nations, will begin operating with Gemini 3.7 Flash. Launched at Google I/O as a personal artificial intelligence agent that acts continuously, carrying out tasks under the user’s direction, Spark becomes even more effective for intellectual activities with this update. The model improves the use of tools for Google Workspace applications, ensuring greater accuracy and quality in workflows that involve complexity and multiple skills.
Thanks to Flash 3.7, Gemini Spark can now turn concepts into actions more efficiently, making it easier to organize files, compose emails, and update status documents.
Security-focused development
The development of Gemini 3.7 Flash considered security as a fundamental pillar. Ongoing efforts are dedicated to strengthening the breadth and strength of Frontier Safety’s security measures. The model arrives with updated protections against inappropriate uses in sensitive areas such as chemical, biological, radiological and nuclear (CBRN) and cyber attacks. These measures, aligned with the bioresilience approach and the cybersecurity program, seek to enable only beneficial use cases.
Gemini 3.7 Flash Availability
Gemini 3.7 Flash is accessible to developers looking to explore agent-based workflows and those starting development on the Gemini API through the Google AI Studio and Android Studio platforms. Companies can use the model through the Gemini Enterprise Agent platform and the Gemini Enterprise application. For the general public, Flash 3.7 is available through Spark, the personal agent in the Gemini app, for Google AI Pro and Ultra plan subscribers in supported countries.