Google unveils multimodal AI "Gemini 1.0" — launches in three sizes, Ultra, Pro, and Nano
Google has unveiled its new-generation large-scale AI model, “Gemini 1.0.” Described by the company as its “largest and most capable” model to date, it stands out for a multimodal design built from the ground up to handle not just text but images, audio, video, and code together. Adopted as the foundation model for Bard, it marked a turning point in Google’s conversational AI strategy.
Details
- Three sizes: “Gemini Ultra” for the most advanced reasoning tasks, “Gemini Pro” for a broad range of uses, and “Gemini Nano,” a lightweight version for on-device processing
- Design philosophy: Trained to be multimodal from the start, rather than combining separate models for text, images, audio, video, and code after the fact
- Rollout: Gemini Pro was made available in Bard the same day. Ultra was said to be planned for release the following year after safety testing
- Benchmarks: Google claimed the model outperformed GPT-4 on major benchmarks, drawing significant attention
- Development: Led primarily by Google DeepMind, announced via a virtual press briefing
What happened next
The Gemini 1.0 announcement made clear Google’s intention to gradually unify the “Bard” brand under “Gemini.” In February 2024, Bard itself was renamed Gemini, and Gemini Advanced (powered by Ultra 1.0) launched — becoming the starting point for the company’s subsequent product rollout.