PeopleofSuperintelligence

Researcher

Kang Guan

Kang Guan is a research and engineering contributor at DeepSeek, the Hangzhou AI lab behind the DeepSeek-V3 and DeepSeek-R1 models. Starting with the DeepSeek LLM paper in January 2024, they coauthored the lab's Mixture-of-Experts models DeepSeek-V2 in May 2024 and DeepSeek-Coder-V2 in June 2024. DeepSeek credited them under Research & Engineering for DeepSeek-V3 in December 2024, and Nature published the DeepSeek-R1 reasoning paper with them among its authors in September 2025. They also coauthored DeepSeek-V3.2 in December 2025 and the DeepSeek-V4 preview in April 2026, whose models support one million tokens of context. As of April 2026, they remain on DeepSeek's research and engineering team.

Now at
DeepSeek

Timeline

  1. January 2024

    The DeepSeek LLM paper

    They coauthor the DeepSeek LLM paper, which DeepSeek releases in January 2024. The project trains open-source models in 7B and 67B configurations on a 2-trillion-token dataset, and DeepSeek LLM 67B surpasses LLaMA-2 70B on code, mathematics and reasoning benchmarks.

  2. May 2024

    Coauthoring DeepSeek-V2

    In May 2024, they coauthor DeepSeek-V2, a Mixture-of-Experts language model with 236B total parameters, 21B of them activated for each token. Compared with DeepSeek 67B, the model saves 42.5% of training costs and reduces the key-value cache by 93.3%.

  3. June 2024

    The DeepSeek-Coder-V2 model

    DeepSeek-Coder-V2, which they coauthor in June 2024, is an open-source code model further pre-trained from a DeepSeek-V2 checkpoint on 6 trillion additional tokens. It expands support from 86 to 338 programming languages and extends the context length to 128K.

  4. December 2024

    Research and engineering on DeepSeek-V3

    DeepSeek credits them under Research & Engineering for DeepSeek-V3, released in December 2024. The Mixture-of-Experts model has 671B total parameters and needs only 2.788M H800 GPU hours for its full training.

  5. January 2025

    The DeepSeek-R1 reasoning model

    In January 2025, they coauthor DeepSeek-R1, a first-generation reasoning model that combines multi-stage training and cold-start data with reinforcement learning. It reaches performance comparable to OpenAI-o1-1217 on reasoning tasks.

  6. September 2025

    DeepSeek-R1 in Nature

    Nature publishes the DeepSeek-R1 paper in September 2025, with Kang Guan among its authors from the DeepSeek-AI Team in Hangzhou, China. The paper shows that pure reinforcement learning can incentivize reasoning in large language models.

  7. December 2025

    DeepSeek-V3.2 and sparse attention

    They coauthor DeepSeek-V3.2 in December 2025, a model built around DeepSeek Sparse Attention for long-context efficiency. Its high-compute variant, DeepSeek-V3.2-Speciale, reaches gold-medal performance at the 2025 International Mathematical Olympiad and International Olympiad in Informatics.

  8. April 2026

    Research and engineering on DeepSeek-V4

    In April 2026, DeepSeek credits them again under Research & Engineering for the DeepSeek-V4 preview, whose two Mixture-of-Experts models support a context length of one million tokens. They remain a member of DeepSeek's research and engineering team.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes DeepSeek. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes DeepSeek. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  3. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staffIncludes DeepSeek. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staff. We test the estimate against 3,913 named model contributors and dated all-staff records across Google DeepMind, OpenAI and Anthropic, plus six additional frontier-grade labs.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes DeepSeek. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Kang Guan yet. Send us a free-licensed photo.

  1. [2401.02954] DeepSeek LLM: Scaling Open-Source Language Models with Longtermism arxiv.org. Checked 20 October 2026.
  2. Paper page - DeepSeek LLM: Scaling Open-Source Language Models with Longtermism huggingface.co. Checked 20 October 2026.
  3. [2405.04434] DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model arxiv.org. Checked 20 October 2026.
  4. [2406.11931] DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence arxiv.org. Checked 20 October 2026.
  5. [2412.19437] DeepSeek-V3 Technical Report arxiv.org. Checked 20 October 2026.
  6. DeepSeek-V3 Technical Report arxiv.org. Checked 20 October 2026.
  7. [2501.12948v1] DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org. Checked 20 October 2026.
  8. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning | Nature nature.com. Checked 20 October 2026.
  9. [2512.02556] DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models arxiv.org. Checked 20 October 2026.
  10. [2606.19348] DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence arxiv.org. Checked 20 October 2026.
  11. DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence arxiv.org. Checked 20 October 2026.