Google has unveiled its latest AI model, Gemini 3.5 Flash, which aims to enhance the efficiency and capability of generative AI tasks. This new version, being rolled out across various Google products, promises to deliver advanced intelligence while optimizing performance to manage complex tasks at scale. A year after introducing its 2.5 version at the I/O event, the advancements in Gemini 3.5 indicate a significant shift in the company’s AI direction.
Improvements in Gemini 3.5 Flash stem from extensive pre-training and insights gained from user feedback. Tulsee Doshi, senior director of product management for Gemini, highlighted that the model can output nearly 300 tokens per second, surpassing earlier versions like Gemini 3.1 Pro, which operates at a much slower rate. This boost in performance could help developers tackle the challenges of creating efficient agentic experiences that require long-running tasks.
Google's emphasis on code generation is particularly noteworthy, aligning with the essential functionalities needed for effective AI agents. Benchmarks such as Terminal Bench and SWE-Bench Pro show substantial improvements with Gemini 3.5 Flash, outpacing older Flash models and demonstrating competitive performance against OpenAI’s GPT 5.5. These advancements may position Google as a strong contender in the AI sector, especially as generative AI increasingly demands more efficient solutions to stay economically viable.
Efficiency and Performance Gains
One major hurdle in deploying generative AI for agentic workflows is the interface design that typically caters to human users. Doshi pointed out the complexity of tasks like user interface control, which require models to navigate pages and execute multiple actions effectively. He noted, “Certain things like UI control are expensive to do because the model has to search the page, it has to know where to click, it has to act through multiple steps.” Nevertheless, he expressed confidence that Gemini 3.5 Flash can handle these tasks well due to its optimal mix of quality and cost.
Internally, Google has already started using Gemini 3.5 Flash, and early metrics show a significant boost in coding performance among its employees. Doshi stated, “We have a set of internal metrics we’ve been evaluating that measures how Googlers code… And you can see a massive, massive jump between where 3.1 Pro was and where 3.5 Flash is.” This internal rollout indicates that the model is not just theoretical; it is already making a noticeable impact on productivity.
Future Developments
Alongside the launch of Gemini 3.5 Flash, Google is upgrading its Antigravity IDE to version 2.0, which will support multiple parallel workflows generated by the new model. This enhancement allows for the creation of sub-agents within the IDE, further leveraging the efficiency of Gemini 3.5 Flash. As Doshi explained, the ability to handle multiple workflows simultaneously is a direct outcome of the new model's efficiency in generating tokens.
As generative AI continues to advance, Gemini 3.5 Flash marks a significant milestone for Google. The combination of increased output speed and improved interface usability may enable more sophisticated applications of AI agents. With ongoing development expected in future iterations, such as the forthcoming 3.5 Pro, the competitive landscape in AI is likely to heat up, pushing other major players to innovate or risk falling behind.
