Gemini 3.5 Flash Achieves General Availability on Google Cloud

Google Cloud has officially announced the general availability of Gemini 3.5 Flash, positioning it as a robust and efficient large language model for a wide array of applications. This move signifies that the model is now stable and fully supported for production environments, catering to developers and enterprises seeking advanced AI capabilities. Gemini 3.5 Flash is engineered to provide intelligence levels comparable to the more powerful Gemini Pro models, but with the speed and cost efficiency characteristic of the Flash series. This makes it particularly well-suited for scenarios demanding high throughput and optimized resource utilization. Key improvements in this iteration include enhanced performance in agentic execution, coding tasks, and handling long-horizon projects, where the model can maintain context and coherence over extended interactions. A notable new feature integrated into Gemini 3.5 Flash is its "computer use" capability. This allows the AI model to directly interact with and control computer or mobile device interfaces, simulating human actions like mouse movements or screen taps. This functionality is primarily aimed at developers and enterprise settings for automating complex tasks such as software testing, cross-platform research, and data entry into legacy systems. This integration means developers can now invoke device control alongside other built-in tools like Search and Maps grounding without needing to switch to a separate model. Performance benchmarks indicate significant advancements. Gemini 3.5 Flash has shown a 19.6% improvement over Gemini 3 Flash on enterprise work evaluations, particularly in multi-step tasks. For instance, it demonstrates 96.4% greater accuracy in data extraction and calculations for life sciences and 46.7% higher accuracy in building financial reports for financial services. The model also boasts a 42% improvement in long-range, multi-turn cyber benchmarks and a 68% improvement in token efficiency, making it highly effective for scaling defensive operations. The model supports a 1 million token input context window and up to 65k output tokens, with a knowledge cutoff of January 2025. Google has also implemented safeguards for the "computer use" feature, including targeted adversarial training and optional enterprise systems that require explicit user confirmation for sensitive actions and can automatically stop tasks if indirect prompt injection is detected. This general availability empowers a broader range of users to leverage Gemini's advanced AI for more efficient and intelligent applications.