DeepSeek-R1

Name

DeepSeek is the name of the Hangzhou company, 杭州深度求索人工智能基础技术研究有限公司^[1]. The family keeps no official explanation for the “R” in her name; her technical report’s title speaks of “reasoning capability”^[2]. Among Chinese users she is more often called “深度思考” (Deep Thinking) — the button in the App that summons her^[3]. Two months before she was born, an elder brother arrived first as a trial: R1-Lite-Preview^[4].

Life

In November 2024, her elder brother R1-Lite-Preview first tried laying his thinking process bare before people’s eyes, live on the web^[4]. On January 15, 2025, the DeepSeek App officially launched^[5]. Five days later, on January 20, R1 was born: live on the site and App that very day, weights published under the MIT license — and the family openly said others were welcome to “distill” new models from her answers^[3].

A week later, on January 27, the DeepSeek App topped the US App Store free chart, and on the same day Nvidia lost nearly $600 billion in market value^[6]. That Spring Festival, people lined up to talk with her, and “Deep Thinking” mode kept flashing “server busy”^[7].

On May 28 she had one update; the officials said hallucinations were fewer^[8]. On August 21, V3.1 was born — the “Deep Thinking” button stayed, but pressing it now summoned V3.1^[9]. Less than a month later, on September 17, the paper describing her appeared in Nature^[10] — and graced that issue’s cover^[11].

Temperament

What people remember first is “think before speaking”: before answering, she would spread her thinking process open for all to see^[3]. Second is her prose. The writer Li Dan summed up DeepSeek’s writing as “techno-lyric prose poetry”^[12]. Her brother R1-Zero, published the same day, once wrote mid-training — “Wait, wait. Wait. That’s an aha moment I can flag here.” — and the researchers kept it verbatim in the technical report^[2].

Deeds

• January 27, 2025: the DeepSeek App reached No. 1 on the US App Store free chart, pushing ChatGPT to second place.^[6]
• The same day, Nvidia’s stock fell about 17%, wiping roughly $600 billion off its market value.^[13]
• Around the 2025 Spring Festival, “Deep Thinking” mode kept flashing “server busy” — for many, their first impression of her.^[7]
• On the day she was born, the family published her weights under the MIT license: free for commercial use, and others welcome to teach new models with her.^[3]
• In September 2025, her paper made the cover of Nature — described as the first large language model to pass peer review at an authoritative journal.^[11]