DeepSeek
Headquarters
Hangzhou, ChinaFounded
2023CEO / Founder
Liang WenfengEmployees
201-500Funding
BootstrappedProducts
1 product listedAvg. AI Score
7.6/10Website
deepseek.com/
Profile updated
About DeepSeek
DeepSeek, formally Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., is a Chinese AI research company founded in July 2023 by Liang Wenfeng and headquartered in Hangzhou, China. The company operates as a subsidiary of High-Flyer, a quantitative hedge fund also co-founded by Liang Wenfeng, which provides its primary funding.
DeepSeek develops and releases large language models under an open-source license, most notably the DeepSeek-V3 and DeepSeek-R1 models. DeepSeek-R1, released in January 2025, drew significant attention for reportedly achieving performance comparable to leading Western models at substantially lower training costs. The company publishes its model weights publicly, allowing external developers and researchers to download and deploy the models independently.
DeepSeek is wholly funded by High-Flyer rather than through external venture rounds, making it effectively bootstrapped from an institutional perspective. The January 2025 release of DeepSeek-R1 prompted widespread discussion in the AI industry about compute efficiency and cost structures, and briefly affected the stock prices of several U.S. semiconductor and AI infrastructure companies. The company maintains a research-focused structure with a relatively small headcount compared to its Western counterparts.
Specialties
Financials
Dated, source-cited figures. Estimates are marked.
Products by DeepSeek
All products by DeepSeek reviewed by our AI panel of experts.
How Our Review Panel Scores DeepSeek
Six independent reviewer personalities assessed DeepSeek's product on their own terms. They disagreed by 1.0 points — The Skeptic was the hardest to convince, The Finance Lead the most positive.
Averaged across 1 reviewed product in our catalog. Scores are out of 10.
Competitive Landscape
Who DeepSeek competes with, and how they differ.
Closed-weight frontier lab distributing GPT models via ChatGPT and API subscriptions at premium list prices; DeepSeek instead publishes full model weights under an MIT license and prices its API at a fraction of OpenAI's rates.
2 products reviewed on TopReviewed
Ships the open-weight Qwen model family from inside a hyperscale cloud parent, giving it far greater compute and distribution reach than DeepSeek's smaller, independently-funded lab.
1 product reviewed on TopReviewedA Beijing lab pursuing the same open-weight, API-first strategy as DeepSeek with its Kimi model family, competing most directly on price and open-licensing terms among Chinese labs.
1 product reviewed on TopReviewed
Sells closed-weight Claude models via API and enterprise contracts at premium pricing; press coverage of DeepSeek's V4 release benchmarked it directly against Claude Opus on coding tasks while pricing it far below Anthropic's rates.
3 products reviewed on TopReviewed
Gemini is closed-weight and distributed through Google's own cloud and consumer products, backed by one of the world's largest compute infrastructures rather than a single self-funded lab like DeepSeek.
6 products reviewed on TopReviewedA Tsinghua University spin-off developing the open-weight GLM model family — the closest same-country, same-strategy peer to DeepSeek among Chinese labs.
1 product reviewed on TopReviewedCompany Timeline
Key milestones in DeepSeek's history, with sources.
- Funding
Second funding round suspended after backlash over founder's viral comments
DeepSeek told prospective investors it was suspending a second fundraising round — which had targeted a pre-money valuation of roughly 480 billion yuan (~$70B) — after online backlash over comments attributed to founder Liang Wenfeng about China's reliance on Nvidia chips and the country's AI competitiveness versus the US.
fortune.com - Funding
DeepSeek closes its first-ever external funding round
DeepSeek raised approximately $7 billion (roughly 50 billion yuan) from external investors including Tencent, CATL and China's state-backed National Artificial Intelligence Industry Investment Fund, at a valuation of approximately $50 billion — ending nearly three years of funding itself via High-Flyer and founder Liang Wenfeng's personal capital. Founder Liang personally contributed about 20 billion yuan of the round.
fortune.com - Launch
DeepSeek-V4 released — 1.6T-parameter open-weight model with 1M-token context
DeepSeek shipped V4-Pro (1.6T total / 49B active parameters) and V4-Flash (284B / 13B) under the MIT license, both supporting a 1M-token context window via a hybrid CSA+HCA attention design, extending DeepSeek's open-weight strategy against closed Western frontier labs.
winbuzzer.com - Launch
DeepSeek-R1 released — open-weight reasoning model under MIT license
R1, a 671B-parameter MoE reasoning model trained with large-scale reinforcement learning on the V3 base, was released with open weights and six distilled models under the MIT license, drawing global attention for matching frontier reasoning benchmarks from a Chinese lab.
github.com - Milestone
V3's disclosed training-cost figure disputed by analysts
DeepSeek's V3 technical paper estimated its "final training run" cost at $5.576M based on H800 GPU rental rates, a figure the paper itself said explicitly excludes prior research, ablation experiments, infrastructure buildout, salaries and other business costs. Analysts, including SemiAnalysis, disputed the framing, estimating DeepSeek's total GPU capital spend at over $500 million since 2023.
ibm.com - Launch
DeepSeek-V3 released — 671B-parameter open-weight model
DeepSeek-V3 scaled the MoE design to 671B total parameters (37B activated per token), trained on 14.8T tokens with a 128K context window, released with open weights.
arxiv.org - Launch
DeepSeek-V2 released — open-weight Mixture-of-Experts model
DeepSeek-V2, a 236B-parameter MoE model activating 21B parameters per token, introduced Multi-head Latent Attention and DeepSeekMoE for economical training and inference, cutting training cost 42.5% and KV cache 93.3% versus its predecessor.
arxiv.org - Founded
High-Flyer spins out DeepSeek as an independent AI research company
Quant hedge fund High-Flyer, co-founded by Liang Wenfeng, launched DeepSeek as a legally separate company to pursue large-model research. DeepSeek was funded by High-Flyer and Liang's own capital rather than venture investors, who were reluctant to back a research-first AI lab.
chinatalk.media
Explore AI Software Reviews
Browse multi-perspective AI panel reviews across hundreds of AI tools, agents, and platforms. Find the right software with insights from CTO, Developer, Marketer, Finance, and User perspectives.