Google's cheapest workhorse model got cheaper and leaner
Google shipped Gemini 3.6 Flash. Alongside it came a faster Flash-Lite for high-volume jobs and a locked-down 'Cyber' security model it's only handing to governments and trusted partners.
What it actually means
The sticker cut is the least interesting part. The model does the same work using ~17% fewer output tokens per job, with fewer reasoning steps and tool calls. Point one real agentic workflow at it this week and watch two things: does quality hold, and does your token count drop. If both, that's free margin.
Source: Google / MarkTechPost · Wednesday 22 July 2026