What is DeepSeek?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
DeepSeek is a model developer and a name used for its AI model family. The name may refer to the developer or its models, rather than the complete environment in which you work. A model is a trained system that interprets information and produces responses; an application supplies the interface and other working capabilities around it.
For example, Goose can connect to a hosted DeepSeek model through OpenRouter. Goose supplied the agent environment, OpenRouter supplied the access route and DeepSeek identified the model being used. This example illustrates a combination of components. It is not a claim that those three names are interchangeable.
Hosted means that the model’s calculations run on computers operated by a provider. The application sends requests over a network and receives responses. An API, or application programming interface, defines how those requests and responses are exchanged. Particular DeepSeek releases also have downloadable model weights: the learned numerical settings needed, with suitable software and hardware, to run the model elsewhere.
Downloadable weights do not mean that running the model is effortless or costless. The computer needs sufficient memory and suitable software. The terms for using a particular release also matter. The model’s origin and the service operating it are separate facts when a third party hosts it.
DeepSeek should also be distinguished from DeepSeek Harness. The latter is an agent working environment that coordinates model calls, files and tools. A tool is a software function such as reading a document or running a command. A model may propose a tool request, but the surrounding application controls and executes it.
For an illustrative comparison, keep the same task and source documents while changing the selected model. Then examine the answer, correction effort and time required. Changing the model, application and source material simultaneously makes interpretation harder.
The useful question is which DeepSeek model is running, where it runs and what the surrounding software allows it to do. A family name is a starting point for identifying the system, not a complete description of its behavior.
Source: DeepSeek — API Documentation.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy