Google releases three new Gemini models — but no 3.5 Pro

What's new

• Gemini 3.6 Flash: primary workhorse with configurable reasoning effort and parallel tool use; 49% on DeepSWE (vs 37% for 3.5 Flash); 17% fewer output tokens; $1.50/$7.50 per MTok
• Gemini 3.5 Flash-Lite: fastest 3.5-class model — 350 output tokens/s; $0.30/$2.50 per MTok; optimized for high-throughput tasks
• Gemini 3.5 Flash Cyber: cybersecurity-specialized model for vulnerability detection and code patching; available only via CodeMender for governments and trusted partners

Why it matters

Gemini 3.6 Flash is the de facto successor to 3.5 Flash with substantially better agentic coding benchmarks. Flash-Lite pushes the cost/performance ratio further; Flash Cyber is the first AI model directly optimized for government security teams. The long-awaited 3.5 Pro is still missing — Google has now missed the deadline three times.

How to try it

All three models are available via Google AI Studio, Android Studio, and Gemini Enterprise Agent Platform. Gemini 3.6 Flash is also available in GitHub Copilot from July 21 for Pro, Pro+, Max, Business, and Enterprise plans.