PeopleofSuperintelligence

Researcher

Xingkai Yu

Also known as 俞星凯, Xing-Kai Yu

Xingkai Yu is a DeepSeek researcher who works on language models and machine-learning systems. He earned undergraduate and master's degrees at Nanjing University. In 2024, he coauthored DeepSeek-V2 and contributed to research and engineering on DeepSeek-V3, then helped implement DeepSeek-R1's reinforcement-learning pipeline in 2025. His 2026 work spans DeepSeek-V4's million-token context, DSpark's speculative decoding, and DeepSeek-V4.1-Flash's 890-byte-per-token global KV-cache footprint. As of October 2026, he focuses on machine-learning systems and large language models at DeepSeek AI.

Now at
DeepSeek
Badges

Timeline

  1. 2021–2025

    Nanjing University degrees

    He earns a B.Sc. in computer science at Nanjing University in July 2021 and enters its M.Sc. program in 2021. He studies in the LAMDA Group in the Department of Computer Science and Technology under Associate Professor Wei Wang; by 2025, he holds undergraduate and master's degrees from NJU.

  2. 2024

    DeepSeek-V2 MoE model

    In 2024, he coauthors DeepSeek-V2, a mixture-of-experts language model with 236B total parameters and 21B activated per token. The model supports a 128K-token context.

  3. 2024

    Research and engineering on V3

    In 2024, he contributes to DeepSeek-V3 research and engineering. The model has 671B total parameters, with 37B activated for each token.

  4. January 2025

    Core contributor to DeepSeek-R1

    In January 2025, he joins DeepSeek-R1's core contributors and helps implement its reinforcement-learning pipeline. He works to optimize system efficiency and address stability issues in large-scale training.

  5. April 2026

    DeepSeek-V4 million-token context

    In April 2026, he coauthors the DeepSeek-V4 report, which describes V4-Pro and V4-Flash versions that support context lengths of one million tokens.

  6. July 2026

    DSpark speculative decoding

    In July 2026, he coauthors DSpark, a speculative-decoding framework deployed in DeepSeek-V4 serving. At matched throughput, DSpark accelerates per-user generation by 60%–85%.

  7. September 2026

    DeepSeek-V4.1-Flash KV-cache compression

    In September 2026, he coauthors DeepSeek-V4.1-Flash. Its designs reduce the global KV-cache footprint to 890 bytes per token.

  8. As of October 2026

    DeepSeek AI research

    As of October 2026, he is a researcher at DeepSeek AI focused on machine-learning systems and large language models.

Career

  1. NowDeepSeek

Education

  1. 2025Nanjing UniversityUndergraduate study
  2. 2025Department of Computer Science and TechnologyM.Sc.

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Xingkai Yu. 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. 16% of AI career moves in the 2020s went from a university to a companyIncludes Xingkai Yu. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  3. 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.
  4. 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.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. Xing-Kai Yu @ LAMDA, NJU-CS lamda.nju.edu.cn. Checked 21 October 2026.
  2. Where Did the DeepSeek Team Go to University? Not in the US. collegetowns.org. Checked 21 October 2026.
  3. DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model arxiv.org. Checked 21 October 2026.
  4. DeepSeek-V3 Technical Report arxiv.org. Checked 21 October 2026.
  5. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org. Checked 21 October 2026.
  6. DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence arxiv.org. Checked 21 October 2026.
  7. DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation | alphaXiv Blog alphaxiv.org. Checked 21 October 2026.
  8. DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation arxiv.org. Checked 21 October 2026.
  9. DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation arxiv.org. Checked 21 October 2026.
  10. DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression arxiv.org. Checked 21 October 2026.
  11. Deepseek researchers | alphaXiv alphaxiv.org. Checked 21 October 2026.