DeepSeek: Open-Source Frontier AI from China — V4-Pro, MIT License
DeepSeek builds open-source frontier models including V4-Pro with 80.6% SWE-bench at a fraction of US competitor costs. Founded 2023 in Hangzhou. Targeting $45–50B valuation.
DeepSeek is headquartered in Hangzhou, China with global teams and API infrastructure serving international users. The company provides worldwide access through cloud infrastructure and partnerships.
Founded: 2023 · HQ: Hangzhou, China · Team: 270-320 · CEO: Liang Wenfeng · Funding: Seeking first external round of $3–4B at ~$45–50B valuation; led by China's Big Fund with Tencent and Alibaba; previously self-funded by High-Flyer hedge fund · Valuation: $45–50 billion (May 2026 est.)
About DeepSeek
DeepSeek was founded in 2023 by Liang Wenfeng, the CEO and co-founder of Hangzhou-based quantitative trading firm High-Flyer. Operating from Hangzhou, China, the company employs approximately 270 researchers and engineers in its core R&D team — a lean structure that reflects its research-lab philosophy. DeepSeek set itself apart early by prioritizing algorithmic efficiency over brute-force compute, demonstrating that frontier AI capability does not require Western-scale GPU clusters. DeepSeek's product lineup has grown rapidly through 2025–2026. The company's R1 reasoning model, released in early 2025, shocked the global AI industry by matching OpenAI o1 performance at a fraction of the cost and triggered a market-wide reassessment of AI infrastructure spending. Subsequent releases — V3, V4, and V4-Pro (late April 2026) — continued this trend, with V4-Pro delivering benchmark performance close to GPT-5.5 and Claude Opus 4.7 at significantly lower API pricing. DeepSeek has also moved into agentic systems: software capable of independently executing complex, multi-step tasks. In May 2026, DeepSeek initiated its first-ever external funding round, seeking $3–4 billion at a valuation of $45–50 billion — a dramatic jump from its prior self-funded status. China's state-backed national AI fund (China Integrated Circuit Industry Investment Fund, "the Big Fund") is in talks to lead the round, with Tencent and Alibaba also participating. The fundraise signals a strategic shift: DeepSeek is transitioning from a capital-light research lab to a state-backed AI powerhouse with resources to scale compute, expand cloud partnerships, and compete aggressively for top talent. Liang Wenfeng's leadership style is distinctive. He met with President Xi Jinping in February 2025 following R1's release — one of only a handful of tech CEOs to do so — and has maintained an exceptionally low public profile since. Internally, the company runs a flat, project-based culture with competitive compensation, resulting in a turnover rate of less than 4% during V4 development. DeepSeek's hiring emphasizes raw capability over experience, with many researchers joining straight from university. Competitively, DeepSeek differentiates through architectural innovation, open-source releases, and cost efficiency. Its models are available via API and open weights, giving the global research community access to evaluate and extend them. V4 was optimized to run on Huawei Ascend chips — a strategic capability that reduces DeepSeek's dependence on restricted Nvidia hardware and positions it as a key pillar of China's domestic AI stack. DeepSeek publishes detailed technical reports alongside each major release, contributing to international understanding of efficient large-model training.
Mission
To advance artificial intelligence through innovative research and open development, creating AI systems that serve global diverse needs while maintaining independence and technological excellence.
Products
- DeepSeek-V2 Models
- DeepSeek-Chat
- DeepSeek-Coder
- DeepSeek-R1 Reasoning
Links
Website · GitHub · Twitter · Blog · Docs
Frequently Asked Questions
What is DeepSeek?
DeepSeek is a Chinese AI research company based in Hangzhou that develops frontier open-weight language and reasoning models. Founded in 2023, it is known for matching GPT-level performance at a fraction of the training and API cost. Its lineup includes the DeepSeek V-series general models and R-series reasoning models.
Who founded DeepSeek?
DeepSeek was founded in 2023 by Liang Wenfeng, who is also CEO of the Chinese quantitative hedge fund High-Flyer. High-Flyer's GPU infrastructure and capital seeded DeepSeek's research. Liang remains the company's leader and primary strategist.
Why is DeepSeek significant?
DeepSeek drew global attention in early 2025 when its DeepSeek-R1 reasoning model rivalled leading US models while reportedly costing far less to train. This challenged the assumption that frontier AI requires billions of dollars in compute. Its low API pricing also pressured incumbents to cut their own prices.
Are DeepSeek's models open source?
Yes, DeepSeek releases open-weight models under permissive licenses, letting developers download, self-host, and fine-tune them. This open approach contrasts sharply with the closed models from OpenAI and Anthropic. It has made DeepSeek popular for cost-sensitive and on-premise deployments.
How much does DeepSeek cost?
DeepSeek is known for aggressively low API pricing, often a fraction of comparable US models, and its open weights can be self-hosted for only the cost of compute. Exact per-token rates vary by model and change over time. This cost advantage is central to its rapid global adoption.
How does DeepSeek achieve lower costs?
DeepSeek uses efficiency techniques such as mixture-of-experts architectures and optimized training pipelines to reach high performance with less compute. Its ties to the High-Flyer hedge fund gave it early access to large GPU clusters. The result is GPT-level quality at dramatically lower training and inference cost.
Where is DeepSeek based?
DeepSeek is headquartered in Hangzhou, China. As a China-based provider, its data handling and availability can be subject to regional regulation, which some Western enterprises weigh before adopting it. Self-hosting the open weights is one way organizations address those concerns.
Who should use DeepSeek?
DeepSeek suits developers and businesses that want strong reasoning and coding performance at minimal cost, and teams that prefer open weights they can self-host. It is less suited to organizations with policies against China-based providers. Its open models are especially popular with researchers and cost-conscious startups.