Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
02:00 · July 21, 2026 · RSS APP - AI Models and Product Updates

We’re introducing new Gemini models, including Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber.
Summary
Google has released three new models in its Gemini Flash line, aimed at developers building production-grade AI agents that require strong token efficiency, low latency, and consistent reliability across multi-step tasks. Gemini 3.6 Flash serves as the primary workhorse, offering gains in coding, knowledge work, and multimodal capabilities while cutting output token consumption by 17 percent relative to 3.5 Flash on the Artificial Analysis Index and by as much as 65 percent on certain DeepSWE evaluations. It also records higher scores on benchmarks such as DeepSWE (49 percent versus 37 percent), MLE Bench (63.9 percent versus 49.7 percent), and OSWorld-Verified (83.0 percent versus 78.4 percent), all at a reduced price of $1.50 per million input tokens and $7.50 per million output tokens. The model ships with strengthened safeguards against chemical, biological, radiological, and nuclear misuse as well as cyber-offense attempts, while limiting unnecessary refusals on legitimate queries.
Gemini 3.5 Flash-Lite targets high-volume, latency-sensitive workloads such as agentic search and document processing. It reaches 350 output tokens per second according to Artificial Analysis and is priced at $0.3 per million input tokens and $2.5 per million output tokens. The model improves markedly over the prior Flash-Lite generation on tasks including Terminal-Bench 2.1 (54 percent versus 31 percent), GDM-MRCR v2 (72.2 percent versus 60.1 percent), and GDPval-AA v2 (1140 versus 642). It also surpasses the earlier 3 Flash model on several agentic and coding evaluations while supporting configurable thinking levels and built-in computer-use tooling.
A third release, Gemini 3.5 Flash Cyber, is a specialized variant fine-tuned for vulnerability detection and remediation. Integrated into the CodeMender agent framework, it coordinates multiple instances to generate consolidated security reports and achieves competitive results on the CyberGym benchmark. Because of its dual-use potential, access is restricted to a limited pilot for governments and trusted partners. The announcement also notes that Gemini 3.5 Pro remains in partner testing, with a broader release planned once ready, and that pre-training for Gemini 4 has already begun.
Why it matters
Direct model update with actionable details on efficiency, cost, benchmarks, and integration for building AI agents. Product teams can evaluate token savings, latency, and coding/multimodal gains for production use. Includes limitations and safety considerations relevant to EU deployment.









