Google DeepMind's natively multimodal model family, announced December 2023, and the model line behind AI Overviews and AI Mode in Google Search.
Gemini is Google DeepMind's multimodal model family, announced in December 2023, and the model line behind AI Overviews and AI Mode in Google Search. It is natively multimodal: text, images, audio and video are handled by one model rather than by encoders bolted onto a text model.
Gemini 1.0 shipped in Ultra, Pro and Nano sizes, the last of which runs on device. Gemini 1.5 Pro introduced a 1 million token context window in February 2024, later extended to 2 million, which is the step that made whole-corpus prompting practical. Later generations split into Flash variants tuned for latency and cost and Pro variants tuned for capability, with explicit reasoning added in the 2.5 generation.
Gemini is reached through the Gemini app, the Gemini API in Google AI Studio, and Vertex AI for enterprise deployments. The same family powers Google's generative search surfaces, which is why its retrieval behaviour decides AI visibility: what Gemini retrieves and cites in AI Mode determines which pages appear in the answer.
