Google has introduced three new Gemini Flash models: 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, designed for efficiency, low latency, and reliability in building AI agents at scale.

3.6 Flash is presented as a workhorse model, offering improved coding, knowledge work, and multimodal performance. It reportedly reduces output token usage by 17% compared to 3.5 Flash and lowers costs, making AI agents more cost-effective. It also shows performance gains in various benchmarks, including Deep SWE and ML Research.

3.5 Flash-Lite is highlighted as the fastest and most cost-effective 3.5-class model, capable of delivering 350 output tokens per second. It is optimized for low-latency and high-throughput tasks like agentic search and document processing, offering a strong price-to-performance ratio.

3.5 Flash Cyber, integrated with the Code Mender agent, is a specialized model for cybersecurity applications, focusing on efficiently finding and fixing vulnerabilities. It will be available to governments and trusted partners through a limited-access pilot program. The company also announced that Gemini 3.5 Pro is in testing, with plans for broad availability, and pre-training for Gemini 4 has begun.