Last updated
    DeepSeek
    information technology, artificial intelligence, large language model

    DeepSeek

    China flagChina
    Founded July 2023 Hangzhou CEO: Liang Wenfeng
    社区

    简介

    In July 2023, Liang Wenfeng spun off a research lab from his quantitative hedge fund High-Flyer and called it DeepSeek. The lab was registered as Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. on July 17, 2023. At that point, High-Flyer had already been using AI for all its trading decisions since 2021, and Liang had been stockpiling Nvidia chips since as early as 2021, anticipating the need for serious computing power. DeepSeek operated for eighteen months as a privately funded lab before seeking outside money.

    The company released its first major model, DeepSeek-V2, in June 2024, triggering a price war among Chinese AI companies including ByteDance, Alibaba, Baidu, Tencent, and Zhipu AI. But the real explosion came in January 2025, when DeepSeek-R1 hit the app store. It became the most downloaded free iOS app in the United States, surpassing ChatGPT. The model was trained for under $6 million, a fraction of what competitors spend, and it used a Mixture of Experts architecture with 671 billion total parameters but only 37 billion activated per token. That efficiency, combined with aggressive pricing at $0.55 per million input tokens, sent shockwaves through Silicon Valley and triggered a stock market selloff.

    DeepSeek employs about 160 to 200 people, mostly young graduates from top Chinese universities. Liang Wenfeng himself is deeply involved in the research and prefers hiring local talent over returnees from overseas. The company's approach is open-source, sharing model parameters under permissive licenses, which has made it popular in the developer community and in regions like Africa where language support is expanding. In 2025, DeepSeek raised $7.4 billion in its first external funding round at a valuation of up to $59 billion, with investors including Tencent and CATL. Liang committed up to $2.8 billion of his own money to maintain control, increasing his direct stake from 1% to 34%. The company faces scrutiny over data scraping allegations and national security concerns, but it continues to push forward, with development underway on future versions.

    产品与子品牌

    DeepSeek-V2DeepSeek-V3DeepSeek-R1DeepSeek CoderDeepSeek-MoEDeepSeek-Math

    常见问题

    谁拥有 DeepSeek?
    DeepSeek 由梁文锋创立的中国量化对冲基金 High-Flyer 拥有。在公司首次外部融资中投入 28 亿美元自有资金后,梁文锋通过直接和间接持股,控制着 DeepSeek 约 84.29% 的股份。
    谁是 DeepSeek 的 CEO?
    梁文锋是 DeepSeek 的创始人兼 CEO。他于 1985 年出生在广东省湛江市,拥有浙江大学电子工程学位,此前他曾联合创立了使用人工智能进行量化交易的对冲基金 High-Flyer。
    DeepSeek 值多少钱?
    DeepSeek 在 2025 年的 74 亿美元融资轮中估值高达 590 亿美元。据报道,该公司盈利能力强,其高效的模型架构将成本降低到标准方法的 5-13%,但估值仍然是一个取决于持续表现的动态目标。
    DeepSeek 与 ChatGPT:哪个更好?
    DeepSeek-R1 在性能基准测试中可与 GPT-4 相媲美,但其训练成本不到 600 万美元,而微软的 AI 基础设施投资为 800 亿美元。DeepSeek 的定价明显更低,为每百万输入 token 0.55 美元,而 ChatGPT 拥有更广泛的全球认知度和企业采用率。
    DeepSeek 还在运营吗?
    是的,DeepSeek 正在积极运营并不断发展。该公司在首次外部融资中筹集了 74 亿美元,持续发布新模型,并通过其聊天机器人在美国应用商店排行榜上名列前茅,实现了全球扩张。包括 v4 在内的未来版本正在开发中。

    关键词

    AI大型语言模型开源中国效率混合专家模型DeepSeek-R1

    来源

    公司信息

    成立
    July 2023
    创始人
    Liang Wenfeng
    首席执行官
    Liang Wenfeng
    总部
    Hangzhou
    行业
    information technology, artificial intelligence, large language model
    母公司
    High-Flyer
    创始人
    Liang Wenfeng
    成立
    July 2023
    总部
    Hangzhou
    行业
    information technology, artificial intelligence, large language model
    Parent Company
    High-Flyer
    首席执行官
    Liang Wenfeng

    官方网站

    deepseek.com

    加入社区

    位粉丝正在讨论