Company

DeepSeek

DeepSeek is a Chinese AI company founded by Liang Wenfeng, originally incubated from quantitative trading firm High-Flyer Quant . It emerged from relative obscurity in early 2025 to become a major global competitor to OpenAI, with its R1 reasoning model and V3 foundation model directly challenging ChatGPT, GPT-4o, and o1 on both capability and cost .

The company's technical approach centers on engineering optimization rather than new paradigms: DeepSeek-V3's 671B-parameter Mixture-of-Experts architecture activates only 37B parameters per token, using innovations like Multi-head Latent Attention, Multi-Token Prediction, and FP8 mixed-precision training to slash costs . It sparked a price war in mid-2024 when V2's API pricing hit 1/35th of GPT-4's rates, forcing domestic rivals to follow suit with steep cuts . R1, released in January 2025, was notable as the first open-source "thinking" model—its training methodology disclosed in detail, letting developers replicate and prompt its reasoning process itself .

DeepSeek's open-source strategy has been central to its impact. As Fusion Fund's Lu Zhang noted, it proved open models could surpass closed ones while running on simpler infrastructure, galvanizing global developer interest . By late 2025, however, its growth had cooled: mobile monthly active users fell 22% from peak, web traffic over 40% .

The company raised substantial external funding in 2026, with IDG Capital and Monolith contributing roughly RMB 3 billion each, Loyal Valley Capital and Shixiang Tech around RMB 1.5 billion apiece—drawn from insurers, state capital, and consumer brands like Nongfu Spring . Liang has maintained unusual autonomy, reportedly willing to assist even rival AI firms provided they don't poach talent .

AI-generated — may contain errors, please verify.

DeepSeekCompany
渲染中…
Mentioned in 479 articles

Coverage