Google has just officially introduced Gemini 3, a new generation AI model, marking a big step forward in reasoning ability and agent features.
Gemini 3 has upgraded reasoning capabilities (Photo: ST).
Gemini 3 builds on Gemini 1's multimodal foundation and Gemini 2's agent mindset, focusing on understanding deep context, capturing subtle layers of meaning, and handling complex problems with fewer commands.
Performance and variants of the Gemini 3 Pro are said to outperform the Gemini 2.5 Pro on most important metrics, the model is tweaked to be concise, direct, straight to the point and with minimal user flattery.
Additionally, Google introduced Gemini 3 Deep Think mode (coming soon), which focuses on advanced reasoning for difficult problems, achieving impressive performance on tests like ARC-AGI (reasoning for artificial general intelligence).
Gemini 3 is deeply integrated into the Google ecosystem with dedicated apps to help synthesize information from text, images, videos , and source code. On Google Search, Gemini 3 offers new interactive interface experiences that help visualize complex topics.
Gemini 3 supports "vibe coding" and interactive web interface creation.
In particular, Google launched Google Antigravity - a new agent programming platform. Here, AI is not just a support tool but also plays the role of an active partner, able to plan, write and test code from start to finish in a specialized IDE (integrated development environment).
Long-term planning capabilities have also improved significantly. AI can perform multi-step processes like automatically sorting Gmail or creating detailed travel itineraries under the user's supervision.
The Gemini 3 Deep Think mode alone is currently in the final safety testing phase and will be released to users soon.