Google launches Gemini 3, which is less flattering and more insightful | Infinium-tech

Google is launching Gemini 3. The latest suite of AI models is certainly the best yet, and replaces Gemini 2.5, which was introduced at Google I/O in May.

Gemini 3 is also being launched in AI mode in Google Search (for Google AI Pro and AI Ultra subscribers) and in the Gemini app for everyone. Google CEO Sundar Pichai says it’s “much better at detecting the context and intent behind your request, so you can get what you need with fewer prompts”.

Google says Gemini 3 is “the world’s best model for multimodal understanding”, providing “rich visualizations and deep interactivity”. Gemini 3 releases start with Gemini 3 Pro, “in preview”. This will be followed by Gemini 3 DeepThink, which is currently being tested by “security testers”. Once they’re done, it will be available to Google AI Ultra customers.

The Gemini 3 Pro “significantly outperforms” the Gemini 2.5 Pro on every major AI benchmark. According to Google, it “brings a new level of depth and nuance to every conversation”, with responses that are “smart, concise and direct, trading clichés and sycophants for real insights – telling you what you need to hear, not just what you want to hear”. The Gemini 3 performs better than the Gemini 3 Pro in the Deep Think benchmark, but of course the tradeoff is that its responses take longer.

Gemini 3 is better at helping you learn anything, because it was built from the ground up to “instinctively synthesize information about any topic across multiple modalities, including text, images, video, audio, and code.”

Google says Gemini 3 is its most secure model yet, whatever that means, and that it “has gone through the most comprehensive set of security assessments of any Google AI model to date”. This reflects “reduced sycophancy, increased resistance to instant injection, and improved protection against abuse through cyber-attacks”.

Source