DeepSeek

What DeepSeek does

DeepSeek is a Hangzhou-based company that builds large language models, the AI systems that generate text and code. It grew out of High-Flyer Capital Management, a Chinese quantitative hedge fund co-founded by Liang Wenfeng, which started DeepSeek as a research lab in 2023. It runs a chatbot app and sells access to its models through an API, an interface that other software can call. It also publishes model weights, the trained parameters that let others run a model themselves.

On April 24, 2026 DeepSeek released preview versions of V4-Pro and V4-Flash, each able to take in 1 million tokens (units of text) at once. On September 10, 2026 it released V4.1-Flash, which it describes as a 552-billion-parameter model that can also interpret images. The release note said V4-Pro would be phased out. But DeepSeek's change log for that day says its API service will continue. Performance figures published with these releases are the company's own.

Why it matters

DeepSeek drew wide attention in January 2025, when it released its R1 reasoning model and its app topped Apple's App Store chart. TechCrunch reported that its rise at least partly caused an 18% fall in Nvidia's share price that month. It said the rise led Wall Street analysts to question whether demand for AI chips would last.

Reuters reported that V4 was adapted for Huawei chip technology and that Huawei said its chips were used in some of the model's training. On September 30, 2026 DeepSeek said it had partnered with Huawei on open-source programming tools for Huawei's Ascend AI chips.

Recent developments

Liang's hedge fund financed DeepSeek until 2026, when it first sought outside investors. In June 2026 it raised about $7.4 billion at a post-money valuation above $50 billion, with Tencent and battery maker CATL becoming its largest external shareholders, Reuters reported. The same report, citing two sources, said DeepSeek had hired CITIC Securities to prepare a listing on Shanghai's STAR Market. A second round began in July 2026. On October 6, 2026 Reuters reported, citing one source, that it was set to raise more than 80 billion yuan, or about $11.9 billion.

On February 23, 2026 Anthropic said DeepSeek and two other Chinese labs had used about 24,000 fraudulent accounts to extract capabilities from its Claude models, a practice known as distillation. It attributed more than 150,000 exchanges to DeepSeek. On September 8, 2026 US officials accused six Chinese companies, including DeepSeek, of distilling American AI models. China's commerce ministry said the allegations lacked a factual basis. DeepSeek has said it has not intentionally used synthetic data generated by OpenAI, Reuters reported in April 2026.

A September 2025 evaluation by the US government's Center for AI Standards and Innovation found that DeepSeek's R1-0528 model answered 94% of overtly malicious requests under a common jailbreaking technique. US reference models answered 8%, it found. It also found DeepSeek models repeated four times as many inaccurate and misleading Chinese Communist Party narratives. The models tested predate V4. Reuters reported in April 2026 that many Western and some Asian governments had barred officials from using DeepSeek, citing data privacy. DeepSeek's privacy policy says it stores personal data in China.

What to watch

• Whether the second funding round closes in October 2026, as two sources told Reuters they expected.
• The timing and size of a STAR Market listing, both undecided as of September 2026, Reuters reported.
• A V4.1-Pro model, which DeepSeek's September 2026 announcement mentions as a future release.
• Independent testing of the V4 family, and any US action following the distillation allegations.