Gemini
Google DeepMind's family of AI models and assistant, designed from the start to work across text, images, audio, video and code.
Google DeepMind, 2023
Instead of a reading brain with eyes glued on later, it learned to read, look and listen at the same time from the beginning.
Where it fits
+All relationships as a list
Is a kind of / part of
Uses / built on
How it works
Google announced Gemini on 6 December 2023. It came from Google DeepMind, the unit formed earlier that year by merging Google Brain and DeepMind. Model versions change frequently, so this page sticks to durable facts.
The defining design choice is being natively multimodal. Many earlier vision-language models took a text-only large language model and attached an image encoder afterwards. Google described Gemini as pre-trained from the start on different modalities and then fine-tuned with more multimodal data, so text, images, audio and video share one model from day one.
Like other assistants, it is a Transformer post-trained with human feedback (see RLHF) to follow instructions.
Everyday use cases
- Case 01
How you've probably met it
Through the Gemini app and assistant, or built into Google products such as search, email and documents.
Strengths, weaknesses and when to use it
Strengths
- Handles mixed inputs (text, images, audio, video) in one model.
Weaknesses
- Like all large language models, it can state incorrect things confidently.
When to use itFor conversational help with tasks that mix text, images and other media.
Go deeper
Sources
- [1]Google, Introducing Gemini (December 2023)
- [2]Gemini Team, Gemini: A Family of Highly Capable Multimodal Models (2023)
Facts checked as of 2026-10-12