From vibe coding to faster models: what’s new in Google’s Gemini update

Just ahead of the holiday season, Google has rolled out a fresh set of updates to Gemini, its artificial intelligence (AI) -powered assistant, underscoring the tech giant’s push to keep pace in a fast-moving AI race.

The new features range from vibe coding tools that enable users to build apps without having to write code to faster models and more conversational search and translation capabilities.

Euronews Next takes a look at the key updates.

Vibe coding

It’s now possible to build an AI-powered app directly on Gemini.

Google said this week it's integrating Opal, its tool for building mini AI apps known as Gems, directly into its web browser.

The tool, introduced in July, develops the blueprint for a multi-step app that works by chaining together prompts, AI model calls, and other tools without a line of code, the company said.

If users want to tweak the prompt they used to build their app, add a new feature or tool, Google said that’s all possible by just describing the change they want to make.

Now, users have a visual editor integrated on Gemini that notes the steps and changes made to the apps. It also lets users preview their apps before sharing them with others who have Google accounts.

​The visual editor also lets users upload images or video links that Opal can use as a reference when building an app.

Opal is one example of a vibe coding tool, an emerging field where AI can generate apps from plain language prompts instead of using code.

Some AI companies, such as OpenAI and Anthropic, released their own versions earlier this year, while Swedish vibe coding startup Lovable raised one of Europe’s biggest early-stage funding of $200 million (€170 million) in its first year of operations.

A faster, cheaper Gemini

A Gemini model that delivers smarts and speed is Google’s tagline for Gemini 3 Flash.

The company said the new Flash model has PhD-level reasoning skills paired with intelligence that keeps up “with the speed of your thoughts”.

It has two modes: a “fast” mode to answer questions quickly and a “thinking” mode that will solve complex problems.

Gemini 3 Flash is also much better than older models at understanding videos and images, and potentially turning that content into a “helpful and actionable plan” in a few seconds.

The company said it is also better than older models at understanding the nuances of a user question and can provide “comprehensive responses that are visually digestible,” by using real-time information from across the internet.

The release of Gemini 3 Flash comes a few weeks after Google released Gemini 3 Pro, the company’s “most powerful” model for vibe coding and autonomous AI agents, and Gemini 3 Deep Think, a mode that tackles complex math, science, and logic problems.

Google says Gemini 3 Flash’s performance “rivals larger frontier models” on the industry’s benchmark tests like the GPQA Diamond, a series of complex questions in biology, physics, and chemistry, and outperforms Gemini 2.5 Flash, an earlier Flash model releasedin June.

Improvements in translation, search

Google also made it easier to chat directly with its Search and Translation features for users in the United States.

Improvements to a Live With Search feature mean that users can now have a back-and-forth voice conversation to get real-time help from the search engine.

The AI-powered voice that answers the query will be “more fluid and expressive than ever before,” the company said in its release.

Last week, Google announced that Gemini is now integrated into the company’s search and translate apps to give users “much smarter, more natural and accurate text translations” while using search.

The company also released a beta version of live translation that will bring “real-time, natural-sounding” translations through headlines.

​One of the improvements to Translate is for it to understand idioms or phrases such as “steal my thunder,” and give a more natural translation for that instead of doing it word-by-word.

It will currently work for users in the United States and India when translating English to over 20 languages, including Spanish, Hindi, Japanese and German.