Google released three new versions of its Gemini AI this week. What they do is great, but greater and closer to the story than that is what they cost.
Every time a company builds a product using AI, they pay per word the AI reads and per word it writes. Those per-word prices have been a barrier between ideation of products and their actual building, especially for people working with Nigerian, not Silicon Valley budgets. This week, Google made those prices cheaper across the board, while still making the models themselves better.
Major Highlights
- Google released two new AI models developers can use immediately: 3.6 Flash, built for everyday tasks like coding and analysis, and a cheaper, faster version called 3.5 Flash-Lite, built for high-volume, simple tasks. Both are cheaper and use fewer resources than the versions they replace, while performing better.
- A third model, 3.5 Flash Cyber, is built specifically to find and fix security holes in software. Google isn’t opening this one to the public; it’s going exclusively to governments and vetted partners through a security tool called CodeMender.
- Both new public models are already live through Google’s developer platforms, with 3.6 Flash also built into Google’s enterprise tools.
- Google confirmed it’s already deep into training its next major model, Gemini 4 — meaning today’s cutting-edge AI will likely be considered basic within a year or two.
- The models ship with stronger built-in safeguards against misuse in high-risk areas like chemical, biological, and cyber threats.
