Google releases optimized Gemini models for scalable AI agent systems

Google announced the launch of three new Gemini models designed to enhance efficiency and reduce costs in developing production-grade AI agent systems. Gemini 3.6 Flash represents a significant advancement over its 3.5 counterpart, consuming 17% fewer output tokens while delivering stronger performance on coding and knowledge-intensive tasks. Priced at $1.50 per million input tokens and $7.50 per million output tokens, the model makes agentic applications more economical. The company introduced Gemini 3.5 Flash-Lite as a high-throughput option built for minimal-latency applications. Processing 350 output tokens per second, this model outperforms earlier Flash versions across multiple benchmarks, including code execution and long-context analysis. At $0.30 per million input tokens, it delivers an attractive cost-to-performance ratio for large-scale workloads. Google also unveiled Gemini 3.5 Flash Cyber, a specialized model for detecting and fixing cybersecurity vulnerabilities. Due to dual-use considerations, access will be restricted to governments and authorized security partners through a limited pilot program. The company continues advancing its roadmap with Gemini 3.5 Pro in partner testing and more ambitious Gemini 4 development underway.