TL;DR: Gemini is Google's family of large language models, developed by Google DeepMind and built to be multimodal from the start. The weights are closed - developers use the Gemini API or Vertex AI, while everyone else meets the family inside Search, Workspace, and Android. Its signature strengths are native multimodality, very long context, and deep integration with Google's ecosystem.
What it is and how you use it
Gemini is made by Google DeepMind, the research arm formed when Google Brain and DeepMind merged. Unlike families that added image or audio support later, Gemini was trained as a multimodal family from the beginning: the same models take text, images, audio, and video as input rather than bolting on separate encoders. Like its peers, it ships in tiers - a flagship tier for the hardest problems and lighter tiers tuned for speed and price - refreshed generation by generation.
Two traits define the family in practice. The first is context size: Gemini models have consistently pushed the large end of the context window range, which suits video analysis, large document sets, and whole-repository code work. The second is distribution. Google threads Gemini through its own products - Search's AI answers, Workspace's writing help, Android's assistant - so the family reaches users who never chose a model at all. For a developer, that same integration shows up as tight hooks into Google Cloud storage, data, and identity services.
Access is hosted and the weights are closed. Hobbyists and small teams start with the Gemini API through Google AI Studio, which has a generous free tier; enterprises use Vertex AI, the Google Cloud platform that adds governance, private networking, and model management around the same family. Google does publish an open-weight sibling family, Gemma, for teams that want open weights they can run themselves - a deliberate two-track strategy: Gemini stays hosted, Gemma is downloadable.
Where it typically fits: applications that lean on multimodal input - screenshots, PDFs, audio, video - or that need very long context at a competitive price, and organizations already committed to Google Cloud, where Vertex AI makes Gemini the path of least resistance. As with its frontier peers, the honest way to choose is to put it on the shortlist and test it against your own workload.
Where it sits in the AI stack
Gemini occupies the model layer, always behind a Google-operated endpoint. Your application talks to the API; Google runs the inference on its own TPU-heavy infrastructure:
Key tools and implementations
-
Gemini API and AI Studio
The fastest path in: prototype in the browser, then call the same models with an API key.
-
Vertex AI
Google Cloud's enterprise platform serving the family with governance, quotas, and private networking.
-
Official SDKs
Client libraries for Python, JavaScript, Go, and other languages covering both the consumer API and Vertex.
-
Gemma open-weight siblings
Google's downloadable model line for local runtimes and fine-tuning when hosted Gemini does not fit.
Related entries
- GPT OpenAI's family of closed-weight language models, served through its API, Azure, and the ChatGPT products.
- Claude Anthropic's family of language models, accessed through an API and major cloud platforms rather than downloadable weights.
- Multimodal AI Models that accept and reason over more than text - images, audio, or video - within one system.
- Choosing a model A framework for weighing capability, latency, cost, privacy, and openness when picking a model for a task.