Top news
Investing in the public benefit OfficialAnthropic NewsOct 18Popular Musk Pledges SpaceX Shares to ‘Every Kid in America’ WebBloomberg TechnologyOct 19Popular Websites are closing their doors to AI agents WebThe Verge AIOct 17 This is the new ChatGPT Voice, powered by GPT Live OfficialOpenAI on YouTubeOct 18 How AI giants will reshape global philanthropy WebAxios AIOct 19 Gemini 4 Argon Really Surprised Me WebTwo Minute PapersOct 19 Google Gemini 4 agents just beat Minecraft with no human intervention WebEngadget AIOct 19 Who could it be? 🧐 OfficialGoogle on YouTubeOct 18 Anthropic gives outside AI watchdog “employee-like” access WebArs Technica AIOct 18 What are self-replicating ‘AI worms’, and why is Anthropic worried about them? WebThe Guardian AIOct 18
GlossaryModelsIntermediate

Foundation model

Also written: Foundation Models framework, Generative AI foundation models, SAM3, Segment Anything Model 3, Segment Anything Models, SAM, SAM 3, Muse models

Large AI models trained on vast data, adaptable for many tasks.

In plain terms

A foundation model is a large artificial intelligence model trained on a massive amount of data. Think of it like a highly educated student who has read an enormous library. Because it has learned so much, it can be adapted to perform many different specific jobs, from writing text to identifying images.

For example, a company might train a foundation model on general internet text. Then, they could fine-tune that same model to become a chatbot that answers customer service questions or to help doctors analyze medical scans. The core learning from the vast initial training makes it efficient to adapt for new purposes.

The expert version

Foundation models are large-scale neural networks, often transformers, trained on broad, diverse datasets. Their emergent capabilities allow them to perform a wide range of tasks with minimal or no task-specific training (zero-shot or few-shot learning). The training process is computationally intensive, but the resulting model can be adapted to numerous downstream applications through fine-tuning or prompt engineering, offering significant efficiency gains over training specialized models from scratch. Tradeoffs include high initial training costs and potential for biases inherited from the training data.

Why it matters now

Foundation models are central to current AI development, with companies like Microsoft and Meta focusing on making them more efficient and adaptable for specific applications, such as weather forecasting or advertising. The drive for more capable and specialized models continues, even as researchers explore the potential for smaller, device-native foundation models.

Where it shows up 34 items, newest first
How each company uses it
Google Gemini
Google Gemini uses foundation models for broad health predictions from wearable data.
Meta Meta AI
Meta AI is improving the efficiency of its foundation models for large-scale ad systems.
Microsoft Copilot
Microsoft is extending open foundation models for weather and Earth-system applications.
OpenAI ChatGPT
OpenAI's foundation models are seen as a benchmark for leaps in robotics, though startups need to build beyond them.
Related terms

Large Language ModelTransformerFine-tuningPrompt EngineeringGenerative AIAI Agent

Often confused with
Specialized Model
A model trained for one specific task, unlike a foundation mo
Pre-trained Model
While foundation models are pre-trained, the term 'foundation
Artificial General Intelligence
Foundation models are a step towards more gener
Origin

The term 'foundation model' was popularized by researchers at the Stanford Institute for Human-Centered Artificial Intelligence (HAI) in their 2021 paper 'On the Opportunities and Risks of Foundation Models'. Source ↗

Try this about 5 minutes

Open a chat interface with a major AI like ChatGPT, Google Gemini, or Claude. Ask it to explain a complex topic in simple terms, then ask it to write a short poem about the same topic. This shows how one model can adapt to different tasks based on your instructions.

Try it: Playbook entries that use thisPlaybook →

The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.

Questions people ask

What makes a model a 'foundation' model?
Its large scale, broad training data, and ability to be adapted for many different downstream tasks without starting from scratch.
Are all large AI models foundation models?
Not necessarily. The term emphasizes the model's role as a base for further adaptation across diverse applications, not just its size.
How are foundation models different from regular AI?
Foundation models are much larger and trained on far more data, giving them a broader range of capabilities that can be specialized, unlike older, task-specific AI.

First seen on Subvolts Jun 23, 2022. Updated Oct 12, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.