Google rolls out Gemini 3 Flash for everyday tasks with improved reasoning

Gemini 3 Flash replaces Gemini 2.5 Flash in Google Gemini app, bringing faster responses for everyday tasks with Gemini 3 Pro-level reasoning

Google expanded its Gemini 3 artificial intelligence (AI) model lineup with the launch of the Gemini 3 Flash on December 17. According to Google, the Gemini 3 Flash is faster compared to the Gemini 3 Pro, which was launched in November, 2025. Gemini 3 Flash succeeds the Gemini 2.5 Flash model and replaces it as the default model in the Gemini app for Fast and Thinking modes, while retaining all capabilities that its predecessor was able to perform.

Gemini 3 Flash: Details

Google said Gemini 3 Flash delivers higher speed and efficiency without sacrificing intelligence, claiming the model matches or rivals larger frontier systems on advanced reasoning and knowledge benchmarks while outperforming Gemini 2.5 Pro across multiple tests. 

 

According to the company, Gemini 3 Flash achieves results comparable to Gemini 3 Pro on multimodal benchmarks, while using fewer tokens on average than Gemini 2.5 Pro by adjusting how much it “thinks” based on task complexity. Google noted that Gemini 3 Flash runs up to three times faster than Gemini 2.5 Pro.

 

The company said the model is designed to help users understand text, images, video, and audio more quickly, and can be used to analyse visual content, interpret audio, and assist with complex questions. 

Gemini 3 Flash is beginning to roll out globally and is becoming the default model in the Gemini app, while also being introduced in AI Mode in Search.

 

The Gemini 3 Flash can perform a range of tasks, such as creating apps by using Canvas, creating images with the Nano Banana, conducting deep research, and explaining concepts with Guided Learning. It can also help users with shopping by enabling agentic experiences. 

Other updates to Gemini

In related news, earlier in November, Google rolled out two experimental AI-driven interfaces in the Gemini app, allowing the system to adapt how answers were presented based on the user’s prompt. One of these, called Visual Layout, enabled Gemini 3 to generate magazine-style, interactive responses featuring images, structured modules, and tappable elements. The second, Dynamic View, allowed Gemini 3 to build and code a lightweight custom interface in real time, producing scrollable and interactive panels that adjusted as users explored explanations, organised information, or followed step-by-step guidance.

Alongside this, Google introduced Gemini Agent, an experimental tool designed to carry out multi-step tasks across Google services such as Gmail, Calendar, and Drive. The company said the agent could help organise inboxes, draft email replies for user approval, create travel plans from email content, or assist with managing bookings, while leaving final decisions with the user. Google noted that Gemini Agent was being rolled out to Google AI Ultra subscribers in the US.