History of DeepSeek in Timeline

DeepSeek, officially Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., is a Chinese AI company specializing in large language models (LLMs). Founded in July 2023 and based in Hangzhou, Zhejiang, it is owned and funded by High-Flyer, a Chinese hedge fund. Liang Wenfeng, co-founder of High-Flyer, serves as CEO of both companies. DeepSeek launched its DeepSeek-R1 model and corresponding chatbot in January 2025.

February 2016: High-Flyer co-founded

In February 2016, High-Flyer was co-founded by Liang Wenfeng.

October 2016: Stock trading using deep learning model

On 21 October 2016, High-Flyer began stock trading using a GPU-dependent deep learning model, having previously used CPU-based linear models.

2017: AI drives most of High-Flyer's trading

By the end of 2017, most of High-Flyer's trading was driven by AI.

2019: Fire-Flyer 1 Constructed

Fire-Flyer 1 was constructed in 2019 and was retired after 1.5 years of operation.

2019: Fire-Flyer computing cluster construction

In 2019, DeepSeek began constructing its first computing cluster, Fire-Flyer, which cost 200 million yuan and contained 1,100 GPUs interconnected at 200 Gbit/s.

2021: AI exclusivity in trading and Nvidia chips

By 2021, High-Flyer was using AI exclusively in its trading algorithms, often using Nvidia chips.

2021: Construction of Computing cluster Fire-Flyer 2 begins

In 2021, Liang began buying large quantities of Nvidia GPUs, reportedly obtaining 10,000 Nvidia A100 GPUs before the United States restricted chip sales to China, and construction began on computing cluster Fire-Flyer 2 with a budget of 1 billion yuan.

2022: Fire-Flyer 2 GPU Configuration

As of 2022, Fire-Flyer 2 had 5,000 PCIe A100 GPUs in 625 nodes, each containing 8 GPUs. It later incorporated NVLinks and NCCL.

2022: Fire-Flyer 2 Hardware and Architecture

In 2022, Fire-Flyer 2 had 5,000 PCIe A100 GPUs in 625 nodes, each containing 8 GPUs, and exclusively used PCIe instead of the DGX version of A100. Later, NVLinks and NCCL were incorporated to train larger models.

2022: Fire-Flyer 2 capacity usage reported

In 2022, it was reported that Fire-Flyer 2's capacity had been used at over 96%, totaling 56.74 million GPU hours, with 27% used to support scientific computing outside the company.

April 2023: Launch of AGI research lab

On 14 April 2023, High-Flyer announced the launch of an artificial general intelligence (AGI) research lab, focusing on AI tools unrelated to finance.

July 2023: DeepSeek founded

In July 2023, DeepSeek was founded by Liang Wenfeng, the co-founder of High-Flyer, who also serves as the CEO for both companies. The company is based in Hangzhou, Zhejiang.

July 2023: Spin-off into DeepSeek

Two months later, on 17 July 2023, High-Flyer's AGI research lab was spun off into an independent company, DeepSeek, with High-Flyer as its principal investor and backer.

November 2023: Release of DeepSeek Coder and DeepSeek-LLM series

In November 2023, DeepSeek released its first model, DeepSeek Coder, followed by the DeepSeek-LLM series on November 29.

November 2023: Release of DeepSeek-LLM Series

In November 2023, the DeepSeek-LLM series was released with 7B and 67B parameters in both Base and Chat forms. The model code is under the source-available DeepSeek License.

2023: V3 Training Cost Comparison with GPT-4

In 2023, DeepSeek trained V3, a predecessor of R1, for US$6 million, significantly less than OpenAI's GPT-4, which cost US$100 million.

2023: GPT-4 Training Cost Comparison

In 2023, DeepSeek trained its V3 model for significantly less than OpenAI's GPT-4, costing US$6 million compared to GPT-4's US$100 million.

January 2024: Release of DeepSeek-MoE models

In January 2024, DeepSeek released two DeepSeek-MoE models (Base and Chat).

May 2024: Liang Wenfeng's stake in DeepSeek

As of May 2024, Liang Wenfeng personally held an 84% stake in DeepSeek through two shell corporations.

May 2024: DeepSeek-V2 series released

In May 2024, DeepSeek released the DeepSeek-V2 series, including 2 base models (DeepSeek-V2, DeepSeek-V2 Lite) and 2 chatbots (Chat).

September 2024: DeepSeek V2.5 Introduced

In September 2024, DeepSeek V2.5 was introduced.

November 2024: DeepSeek-R1-Lite preview available

On 20 November 2024, the preview of DeepSeek-R1-Lite became available via chat.

2024: DeepSeek-R1-Lite performance comparison

In 2024, DeepSeek claimed that DeepSeek-R1-Lite-Preview exceeded performance of OpenAI's o1 on benchmarks such as AIME and MATH. However, The Wall Street Journal reported that on 15 problems from the 2024 edition of AIME, the o1 model reached a solution faster.

January 2025: Impact of R1 release on AI industry

After the January 2025 release of the R1 model, which offered significantly lower costs than competing models, some investors anticipated a price war in the American AI industry.

January 2025: Chatbot and DeepSeek-R1 model launch

In January 2025, DeepSeek launched its eponymous chatbot alongside the DeepSeek-R1 model.

January 2025: DeepSeek-R1 model release

In January 2025, DeepSeek released the DeepSeek-R1 model under the MIT License.

January 2025: DeepSeek chatbot launch and app store success

On 20 January 2025, DeepSeek launched the DeepSeek chatbot for iOS and Android, and by 27 January, it surpassed ChatGPT as the most downloaded freeware app on the iOS App Store in the United States.

January 2025: New models available under open-source licenses

Since the January 2025 debut of DeepSeek-R1, DeepSeek has made its new models available under free and open-source software licenses, primarily the MIT License.

January 2025: R1 preprint on arXiv

With the release of R1 in January 2025, the DeepSeek team published a preprint on arXiv.

March 2025: DeepSeek-V3-0324 Released

On 24 March 2025, DeepSeek released DeepSeek-V3-0324 under the MIT License.

May 2025: R1 updated instead of R2 release

In early May 2025, R2, the intended successor to R1, was originally planned for release. However, on 28 May 2025, R1 was instead updated to version R1-0528.

May 2025: DeepSeek-R1-0528 Released

On 28 May 2025, DeepSeek released DeepSeek-R1-0528 under the MIT License. The model has been noted for more tightly following official Chinese Communist Party ideology and censorship in its answers to questions than prior models.

August 2025: DeepSeek V3.1 Released

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 and surpasses prior models like V3 and R1 by over 40% on certain benchmarks.

September 2025: DeepSeek V3.1-Terminus and V3.2-Exp Released

In September 2025, DeepSeek V3.1 was updated to V3.1-Terminus on September 22, and V3.2-Exp was released on September 29. V3.2-Exp uses DeepSeek Sparse Attention.

September 2025: R1 publication in Nature

Later, in September 2025, an updated version of R1 was published in Nature.

December 2025: DeepSeek V3.2 Released

On 1 December 2025, DeepSeek-V3.2 was released, alongside a DeepSeek-V3.2-Speciale variant that focused on reasoning.

2025: Fire-Flyer 2 Still in Operation

As of 2025, Fire-Flyer 2 is still in operation.

February 2026: Anthropic Accusation

In February 2026, Anthropic accused DeepSeek of using thousands of fraudulent accounts to generate millions of conversations with Claude to train its own large language models.

April 2026: Funding Round Discussions

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.

April 24, 2026: DeepSeek V4 Series Preview Released

On April 24, 2026, DeepSeek released a preview of its V4 series, including the 1.6-trillion parameter DeepSeek-V4-Pro and the 284-billion parameter DeepSeek-V4-Flash, both featuring a 1-million token context window, under the MIT License.