DeepSeek

From Remilia Wiki
Jump to navigation Jump to search


DeepSeek-Logo
The DeepSeek logo

DeepSeek is a Chinese artificial intelligence company based in Hangzhou, known for releasing powerful open-weight language models at a fraction of the cost of its American rivals. It was founded in July 2023 by Liang Wenfeng and incubated by High-Flyer, a quantitative hedge fund Liang founded in 2015.[1]

V3 and R1

DeepSeek-V3, released in December 2024, is a mixture-of-experts model with "671B total parameters with 37B activated for each token". DeepSeek reported that its final training run cost about $5.6 million in rented computing time, a figure that excludes earlier research and experiments.[2] In January 2025, it released DeepSeek-R1, a reasoning model it said performed "on par with OpenAI-o1", with its code and weights released under the MIT License.[3]

R1's release shook Western markets, which had assumed that frontier AI required enormous spending. On January 27, 2025, Nvidia's stock "plummeted 17%", losing close to $600 billion in market value, the largest one-day loss in US history.[4] The release also intensified debate over US export controls on AI chips; Dario Amodei of Anthropic argued for keeping them in place. A peer-reviewed paper on R1 was published in Nature in September 2025.

Later models

DeepSeek has continued to release open-weight models, including V3.1 (August 2025), V3.2 (December 2025) and the V4 preview (April 2026), whose largest version has 1.6 trillion total parameters and is licensed under the MIT License.[5]

See also

References

  1. ↑ Chen, Caiwei (January 24, 2025). "How a top Chinese AI model overcame US sanctions". MIT Technology Review. Retrieved November 13, 2026.
  2. ↑ December 27, 2024. "DeepSeek-V3 Technical Report". arXiv. Retrieved November 13, 2026.
  3. ↑ January 20, 2025. "DeepSeek-R1 Release". DeepSeek. Retrieved November 13, 2026.
  4. ↑ Subin, Samantha (January 27, 2025). "Nvidia sheds almost $600 billion in market cap, biggest one-day loss in U.S. history". CNBC. Retrieved November 13, 2026.
  5. ↑ April 24, 2026. "DeepSeek-V4 Preview". DeepSeek. Retrieved November 13, 2026.