Back to Lexicon

Deepseek

novice

A Chinese AI research lab known for highly efficient open-source models. Their Deepseek-V3 and reasoning models demonstrated that frontier capabilities can be achieved with significantly less compute than Western labs.

Category: frameworks
companiesopen-sourceefficiency

Overview

Deepseek is a Chinese AI company that shocked the industry with models matching or exceeding Western performance at a fraction of the cost. Their work challenges assumptions about the compute requirements for frontier AI. Deepseek-V3 reportedly trained for under $6 million—compared to hundreds of millions for comparable Western models. Their open-source releases enable researchers worldwide to study efficient training techniques. The company's success has implications for AI geopolitics, open-source development, and the economics of AI research. It demonstrates that algorithmic innovation can compensate for compute limitations.

Key Concepts

Compute Efficiency

Achieving frontier performance with dramatically less training compute.

Open Source Leadership

Releasing powerful models with open weights for research.

Algorithmic Innovation

Novel training techniques that improve sample efficiency.

Related Concepts