Google's June AI drop, translation, video, images, and a model that runs on your laptop

Google rarely ships one thing at a time, and June has been no exception. Over the past couple of weeks the company has rolled out a batch of new models and capabilities across the Gemini family. None of them is a single blockbuster release, but together they sketch out where Google is heading: agents that can operate software, and AI that handles voice, video, and images as naturally as text.

Live translation across 70+ languages

The standout for everyday use is Gemini 3.5 Live Translate, an audio model built for near-real-time speech-to-speech translation. It detects more than 70 languages and is designed to preserve the speaker's intonation rather than producing flat robotic output. It is rolling out through the Gemini Live API, Google AI Studio, and the Google Translate app.

New models for images and video

Two more additions target creative and media work. Nano Banana 2 Lite is Google's fastest and cheapest Gemini image model so far, aimed at high-volume image generation where cost matters more than maximum fidelity. Gemini Omni Flash, now in public preview through the API, is a natively multimodal model built for dynamic video workflows — the sort of thing developers use to analyze, summarize, or generate around video content.

Agents that can use a computer

Gemini 3.5 Flash, the workhorse model Google launched at I/O in May, picked up a Computer Use capability in June. In plain terms, that means agents built on it can look at a screen, reason about what is there, and click, type, and navigate across desktop, mobile, and browser environments. Google is pitching this at long-running automation tasks — the multi-step chores that previously needed a human to babysit them.

Gemma 4 12B: capable AI, no cloud required

Finally, Gemma 4 12B is a new open model sized to run locally on a laptop with 16GB of memory. It has a unified architecture that handles vision and native voice, which is unusual for a model this small. For anyone who cannot send data to the cloud — or just does not want a monthly API bill — this is the most interesting item in the batch.

Why it matters for small business

A few practical takeaways. If you serve customers in multiple languages, live translation that keeps a natural tone is getting close to phone-call quality, and it will show up inside tools you already use. If you produce a lot of product images or short video content, the cost per asset keeps dropping — worth re-checking quotes from any agency billing you at last year's rates. And Gemma 4 running on an ordinary laptop means genuinely private AI is now realistic for businesses handling sensitive client data, no server room required.