Google DeepMind presents Gemini 3
Google DeepMind announced today, June 11, 2026, the release of Gemini 3, its newest artificial intelligence model. The system is capable of processing text, image, audio, and video in real time, marking a significant advance in the field of multimodal reasoning.
Superior benchmark performance
According to information released by DeepMind, Gemini 3 surpasses GPT-4 on complex reasoning benchmarks, including MMLU (Massive Multitask Language Understanding) and GSM8K (Grade School Math 8K). The results position the model as one of the most advanced on the market, although specific score details have not been disclosed.
Energy efficiency
A highlight of the announcement is training efficiency. DeepMind states that the process consumed 25% less energy compared to the previous version of Gemini. The reduction reflects the company's ongoing efforts to make AI more sustainable.
Real-time multimodal capabilities
Gemini 3 sets itself apart by integrating multiple data modalities simultaneously. Unlike models that process each format separately, Gemini 3 combines text, image, audio, and video into a single analysis, enabling more contextual and faster responses. DeepMind demonstrated applications such as real-time video interpretation with automatic captions and object recognition with instant audio description.
Market impact
Experts believe the launch intensifies competition in the AI sector, especially against OpenAI and Anthropic. Real-time multimodal processing capabilities could open new possibilities in areas such as virtual assistants, media analysis, and autonomous systems.
Next steps
DeepMind has not yet announced the public availability date for Gemini 3 but indicated that partnerships with developers and researchers will be announced soon. The company states it will continue to improve the model with a focus on safety and bias reduction.