DeepSeek Glossary

Chinese AI company founded in 2023 and funded by High-Flyer; known for releasing the DeepSeek-V3 and DeepSeek-R1 model weights.

DeepSeek is a Chinese AI company founded in 2023 by Liang Wenfeng and based in Hangzhou. High-Flyer, the quantitative hedge fund also founded by Liang, funded the company. DeepSeek has released downloadable model weights, an API, and a consumer chat service.

Its best-known early releases were DeepSeek-V3, a mixture-of-experts model with 671 billion total parameters and 37 billion activated for each token, and DeepSeek-R1, a January 2025 reasoning model trained through a combination of reinforcement learning and supervised fine-tuning. DeepSeek released the weights for both models and several smaller models distilled from R1.

For this Dictionary, DeepSeek matters technically and strategically. Its releases give operators additional options for Sovereign Compute and demonstrate that a Chinese lab can produce models competitive with leading US systems while distributing the weights. Downloadable weights do not make a 671-billion-parameter model easy or cheap to run; the smaller distilled models make local use more practical.

Sources

• DeepSeek, DeepSeek-V3 model repository: https://github.com/deepseek-ai/DeepSeek-V3
• DeepSeek, DeepSeek-R1 model repository: https://github.com/deepseek-ai/DeepSeek-R1
• Congressional Research Service, DeepSeek and the Race to Develop Artificial Intelligence, July 1, 2025: https://www.congress.gov/crs-product/IF13051

See also

• Reasoning Model
• Sovereign Compute
• Capability Overhang
• Open-Weights Inversion (covered inside Sovereign Compute)

Return to Dictionary All Entries (A–Z) For Students Other Writing Capstone 2.0