Google Korea Blog Link
https://blog.google/intl/ko-kr/company-news/technology/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/
For developers and customers building real-world service environments with AI agents, higher token efficiency, lower latency, and stable performance are crucial. Google's Flash model series is designed to meet these requirements by balancing optimal efficiency and performance, enabling large-scale expansion of agent workflows. Based on Gemini 3.5 Flash, Google introduces the following new Gemini models:
3.6 Flash: This flagship model enhances coding, knowledge-based tasks, and multimodal performance. According to the Artificial Analysis Index, it reduces output token usage by 17% compared to Gemini 3.5 Flash. On some benchmarks like Datacurve's DeepSWE, it achieves up to a 65% reduction in output token usage while also lowering the cost per output token.
3.5 Flash-Lite: This is the fastest and most cost-efficient model among Gemini 3.5 models. It generates 350 output tokens per second according to the Artificial Analysis Index, providing superior performance in agent workflows compared to previous Flash-Lite versions.
3.5 Flash Cyber in CodeMender: Effective implementation of cybersecurity applications requires close integration between models and agent infrastructure. To achieve this, Google combines a new model specialized for cybersecurity with 'CodeMender,' a code security agent, to deliver cutting-edge competitive performance.
In addition to the models announced today, Google is currently conducting tests of Gemini 3.5 Pro with partners and plans to officially release it as soon as preparations are complete. Simultaneously, the Google team is already focusing on developing the next generation of models. Google has embarked on its most ambitious pre-training effort yet and is accelerating the development of Gemini 4, with exciting advancements to come.
https://x.com/GeminiApp/status/2079591659965468743
