Google launches Gemini 3.6 Flash and Cyber models, pulls Earth AI image feature after backlash

Google unveiled a trio of new Gemini models — Gemini 3.6 Flash for speed, 3.5 Flash-Lite for low cost, and 3.5 Flash Cyber for security workflows — and expanded the Gemini API's Managed Agents to support background tasks and remote Model Context Protocol (MCP) connections. The Cyber variant slots into the week's broader security theme, arriving as rivals ship dedicated security models (Microsoft's MAI-Cyber-1-Flash) amid the agent-breach scare.

The Managed Agents update is the developer-facing centerpiece: remote MCP support lets agents call external tools and services over a standardized protocol, and background tasks enable long-running agentic workflows without holding an open session — mechanics aimed squarely at the same agentic-app market Azure and AWS are chasing.

But the launch was marred by a rapid reversal: Google pulled its Earth AI image generation feature (the 'nano-banana 2' Earth scenes) just one day after release following criticism that the AI-generated imagery was misleading. It's a fresh angle on Google's recurring struggle to ship generative-image features without accuracy or safety blowback.

Competitively, Gemini 3.6 Flash is the parity target DeepSeek's V4-Flash-0731 just matched at a fraction of the price, underscoring how quickly open-weight Chinese models are closing on Google's efficiency tier. What to watch: whether the Cyber model gains enterprise traction post-breach-week, and whether Google reworks and relaunches Earth AI imagery or quietly shelves it.