| Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. | |
|---|---|
Native name | 杭州深度求索人工智能基础技术研究有限公司 |
| Type | Private |
| Industry | Information technology Artificial intelligence |
| Founded | 17 July 2023 |
| Founder | |
| Headquarters | Hangzhou, Zhejiang, China |
| Key people |
|
| Products | DeepSeek |
| Owner | High-Flyer |
Number of employees | 396 (2026) |
| Website | deepseek.com |
| Part of a series on |
| Artificial intelligence (AI) |
|---|
| Glossary |
Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., doing business as DeepSeek, is a Chinese artificial intelligence (AI) company that develops open weight large language models (LLMs). Based in Hangzhou, China, DeepSeek is owned and funded by High-Flyer, a Chinese hedge fund.
DeepSeek was founded in July 2023 by Liang Wenfeng, who also serves as CEO. It launched an eponymous chatbot alongside its DeepSeek-R1 model in January 2025.
Models are open-weight; since January 2025 the company offers external users an open license for model parameters, while not offering an open license to training data. The company reportedly recruits AI researchers from top Chinese universities and from other fields to broaden its models' knowledge and capabilities.
DeepSeek is considered closely linked to China's defense industry due to its researchers previous affiliations with and the adoption of its chatbot for non-combat roles by the People's Liberation Army since March 2025.[needs update?] According to The New York Times, dozens of DeepSeek researchers have or have previously had affiliations with People's Liberation Army laboratories and the Seven Sons of National Defence. Since 2025, its chatbot was used in non-combat roles by the People's Liberation Army, including hospitals, the People's Armed Police paramilitary, and national mobilization organizations.
Liang Wenfeng co-founded the company In June 2015, funded by High-Flyer . By 2021, he had started buying large quantities of Nvidia GPUs for an AI project,[clarification needed] reportedly obtaining 10,000 Nvidia A100 GPUs before the United States restricted chip sales to China.[citation needed]
On 14 April 2023, High-Flyer announced the launch of an artificial general intelligence (AGI) research firm, stating that it would focus on developing AI tools unrelated to the firm's financial business. Two months later, on 17 July 2023, that lab was spun off into an independent company. That lab was named DeepSeek, with High-Flyer as its principal investor and backer. Initially, venture capital investors were reluctant to provide funding, as they considered it unlikely that the venture would be able to quickly generate an exit (an ownership stake in an investment).[clarification needed]
In September 2026, Bloomberg reported that DeepSeek was planning to deploy a data center in Inner Mongolia hosting 160,000 Huawei Ascend 950DT chips.
As of May 2024, Liang personally held an 84% stake in DeepSeek through two shell corporations.[needs update?] In February 2026, Anthropic accused DeepSeek of using thousands of fraudulent accounts to generate millions of conversations with Claude to train its own LLMs via distillation. In April 2026, investors began speaking with DeepSeek for a $300 million funding round, which would bring DeepSeek to a total valuation of $10 billion.[needs update] DeepSeek completed a May 2026 Series A round receiving US$7 billion, reaching a post-money valuation of US$52 billion. In July 2026, Bloomberg and the Financial Times reported the company had begun preparations for an IPO, listing as soon as 2027. That month, it began a separate discussion targeting a US$70 billion pre-money valuation.

DeepSeek operates with a relaxed, academic research-style workplace culture that deliberately avoids China's "996" overtime practices. Liang structured the company to eliminate rigid key performance indicators (KPIs) and mandatory overtime, allowing significant time for unstructured research and creative exploration.
DeepSeek's hiring approach emphasizes skills over lengthy work experience, resulting in many employees joining just after completing university. Likewise, the company recruits individuals without computer science backgrounds to expand the range of expertise incorporated into the models, for instance in poetry or advanced mathematics.
DeepSeek initially stated that it focuses on research and does not have immediate plans for commercialization. This posture allowed it to skirt certain provisions of China's AI regulations aimed at consumer-facing technologies.[needs update][original research?] Its use of open weights and MIT licensiing also distinguish it. US chip export restrictions led DeepSeek to prioritize training and inferencing efficiency over model capability, leveraging available compute and reducing energy consumption.: 19 US vendors such as Anthropic, OpenAI, Google, and SpaceX instead are proprietary and compete primarily on model capability instead of performance.
DeepSeek operates in Africa, offering more affordable and less power-hungry AI solutions. The company has bolstered African language models and generated startups, for example in Nairobi.[citation needed] Along with Huawei's storage and cloud computing services, it aided the technology scene in sub-Saharan Africa. DeepSeek offers local data sovereignty and more flexibility compared to western AI platforms.
High-Flyer/DeepSeek operated at least two primary computing clusters: Fire-Flyer (萤火一号) and Fire-Flyer 2 (萤火二号). Fire-Flyer 1 was constructed in 2019 and was retired after 1.5 years. Fire-Flyer 2 was still in operation as of 2025. Fire-Flyer 2 consists of co-designed software and hardware. On the hardware side, Nvidia GPUs use 200 Gbps interconnects. The cluster is divided into two "zones", and the platform supports cross-zone tasks. The network topology was two fat trees, chosen for high bisection bandwidth. On the software side are:
3FS (Fire-Flyer File System): A distributed parallel file system designed for asynchronous random reads. It uses Direct I/O and RDMA Read. In contrast to standard Buffered I/O, Direct I/O does not cache data. Caching is useless in this case, since each piece of data read is random and is not reused.hfreduce: Library for asynchronous communication, originally designed to replace Nvidia Collective Communication Library (NCCL). It is mainly used for allreduce, especially of gradients during backpropagation.[citation needed] It runs asynchronously on the CPU to avoid blocking kernels on the GPU. It uses two-tree broadcast like NCCL.hfai.nn: Software library of commonly used operators for neural network training, similar to torch.nn in PyTorch.[citation needed]HaiScale Distributed Data Parallel (DDP): Parallel training library that implements various forms of parallelism such as data parallelism (DP), pipeline parallelism (PP), tensor parallelism (TP), experts parallelism (EP), fully sharded data parallel (FSDP) and zero redundancy optimizer (ZeRO). It is similar to PyTorch DDP, which uses NCCL on the backend.HAI Platform: Various applications such as task scheduling, fault handling, and disaster recovery.As of 2022, Fire-Flyer 2 had 5,000 PCIe A100 GPUs in 625 nodes, each containing 8 GPUs.
This section needs expansion. You can help by adding missing information. (August 2026) |
DeepSeek Harness is a company-built n open-source AI agent harness. It provides an environment for building and running AI agents, with components including models, tools, skills, sessions, sandboxes, storage, scheduling, and user interfaces. The software is distributed under the MIT License. As of August 2026, DeepSeek Harness is available as a developer preview.
DeepSeek Harness uses an "everything-is-a-plugin" design philosophy powered by the Cordis framework. This allows major agent components to be added, removed, or reloaded while the software is running. The modular design enables developers to switch between third-party providers or alternative model configurations without changing the rest of the system.
On 20 January 2025, DeepSeek launched DeepSeek-R1, which was free for iOS and Android. By 27 January, DeepSeek surpassed ChatGPT as the most downloaded freeware app on the iOS App Store in the United States, triggering an 18% drop in Nvidia's share price.
On 28 May 2025, DeepSeek released a minor update, DeepSeek-R1-0528 under the MIT License. Its scores exceeded both DeepSeek R1 (January) and DeepSeek V3-0324 in certain benchmarks.
On 21 August 2025, DeepSeek released DeepSeek V3.1 under the MIT License. This model features a hybrid architecture with thinking and non-thinking modes. It surpassedV3 and R1, by over 40% on benchmarks such as SWE-bench and Terminal-bench. It was updated to V3.1-Terminus on 22 September, fixing multiple language artifacts and improved intelligence.
DeepSeek V3.2-Exp was released on 29 September 2025. It uses DeepSeek Sparse Attention, a more efficient attention mechanism based on research published in February. It was based on DeepSeek V3.1 Terminus. DeepSeek-V3.2 was released on 1 December 2025, alongside DeepSeek-V3.2-Speciale, a variant that focused on reasoning.
This section needs to be updated. The reason given is: info about V4.1-Flash-1017 is needed. (October 2026) |
On 24 April 2026, DeepSeek released a preview of its V4 series, including 284-billion parameter DeepSeek-V4-Flash and 1.6-trillion parameter DeepSeek-V4-Pro. Both featured a one million token context window. V4 was adopted by semiconductor manufacturers such as Huawei and Cambricon Technologies. The official versions of DeepSeek V4-Flash and V4-Pro were released on 31 July and 13 August, respectively.
V4.1-Flash was released on 10 September, offering a different architecture. It used a causal encoder-decoder, while V4 used only a causal decoder. V4.1-Flash had lower memory usage compared to previous versions, according to DeepSeek. It was natively multimodal.
| Major versions | Release date | Status | Major variants | License | Remarks |
|---|---|---|---|---|---|
| DeepSeek-Coder | November 2, 2023 | Discontinued | Base (pretrained) Instruct (with instruction-finetuned) |
Open-weights (DeepSeek) | Uses the same framework as DeepSeek-LLM. |
| DeepSeek-LLM | November 29, 2023 | Discontinued | Base Chat (with SFT) |
Uses the same architecture and training algorithm as the original Llama model, with a different learning rate scheduler. | |
| DeepSeek-MoE | January 9, 2024 | Discontinued | Base Chat |
Developed a variant of MoE. | |
| DeepSeek-Math | April 2024 | Discontinued | Base | Initialized with DeepSeek-Coder-Base-v1.5 | |
| Instruct (with SFT) | |||||
| RL (using a process reward model) | Developed Group Relative Policy Optimization (GRPO), a variant of Proximal Policy Optimization (PPO). | ||||
| DeepSeek-V2 | May 2024 | Discontinued | DeepSeek-V2, DeepSeek-V2-Chat DeepSeek-V2-Lite, DeepSeek-V2-Lite-Chat DeepSeek-Coder-V2 DeepSeek-V2.5 |
Developed multi-head latent attention (MLA). Also used MoE. Implemented KV caching. | |
| DeepSeek-V3 | December 2024 | Active | DeepSeek-V3-Base DeepSeek-V3 (a chat model) |
The architecture is the same as V2. The training algorithm is the same except with improvements to MoE load balancing and multi-token decoding. Updated on 24 March 2025. | |
| DeepSeek-Prover-V2 | May 1, 2025 | Active | DeepSeek-Prover-V2-671B DeepSeek-Prover-V2-7B |
||
| DeepSeek-VL2 | December 13, 2024 | Active | |||
| DeepSeek-R1 | November 20, 2024 | Active | DeepSeek-R1-Lite-Preview | Proprietary | Preview version, only accessed through API and a chat interface. |
| January 20, 2025 | Active | DeepSeek-R1 DeepSeek-R1-Zero |
MIT | Initialized from DeepSeek-V3-Base and sharing the V3 architecture. | |
| Distilled models | Initialized from other models, such as Llama, Qwen, etc. Distilled from data synthesized by R1 and R1-Zero. | ||||
| May 28, 2025 | Active | DeepSeek-R1-0528 | |||
| DeepSeek-V3.1 | August 21, 2025 | Active | DeepSeek-V3.1-Base DeepSeek-V3.1 (a chat model) |
Hybrid architecture (thinking and non-thinking modes available). Trained on over 800B additional tokens on top of V3. | |
| September 22, 2025 | Active | DeepSeek-V3.1-Terminus | Reducing instances of mixed Chinese-English text and occasional abnormal characters on top of V3.1. | ||
| DeepSeek-Math-V2 | November 27, 2025 | Active | Apache 2.0 | ||
| DeepSeek-V3.2 | December 1, 2025 | Active | DeepSeek-V3.2 DeepSeek-V3.2-Speciale |
MIT | |
| DeepSeek-V4 | April 24, 2026 | Active | V4-Pro, V4-Flash | Preview release on 24 April. Official release of V4-Flash on 31 July, and V4-Pro on 13 August | |
| DeepSeek-V4.1 | September 10, 2026 | Active | V4.1-Flash | Uses a very different architecture than V4. Natively multimodal. |
In the United States, the National Defense Authorization Act for Fiscal Year 2026 instructed the United States Secretary of Defense and the Director of National Intelligence to remove and exclude artificial intelligence developed by DeepSeek or owned by HighFlyer from devices operated by the United States Department of Defense and United States Intelligence Community or their contractors. Artificial intelligence by DeepSeek is allowed only with explicit permissions for research, training, and evaluation or military activities supporting national security functions such as counterterrorism or counterintelligence.
In Australia, the Department of Home Affairs issued a government-wide directive "to prevent the use or installation of DeepSeek products, applications and web services and where found remove all existing instances of DeepSeek products, applications and web services from all Australian Government systems and devices".