PeopleofSuperintelligence

Researcher

Xuecheng Su

Xuecheng Su is a researcher whose credited work includes DeepSeek AI model papers on scaling, training systems and reasoning. He coauthored the January 2024 DeepSeek LLM scaling paper and DeepSeek-V2 work on a 236-billion-parameter mixture-of-experts model with 21 billion active per token. At SC24 in November 2024, Su was listed with DeepSeek AI for the Fire-Flyer AI-HPC paper, which reports that Fire-Flyer 2 trained on 10,000 PCIe A100 GPUs at performance approximating DGX-A100 while reducing costs by half and energy use by 40%. DeepSeek-V3 scaled to 671 billion parameters; R1 explored pure reinforcement learning for reasoning, and V3.2 added sparse attention for long contexts. DeepSeek-V4 introduced Pro and Flash models with one-million-token contexts, and as of October 2026, Su is a listed author on V4.1-Flash, whose report reduces the global KV cache footprint to 890 bytes per token.

Timeline

  1. January 2024

    DeepSeek LLM scaling

    In January 2024, Su coauthors DeepSeek LLM, a study of scaling laws for open-source language models in 7B and 67B configurations.

  2. May 2024

    DeepSeek-V2's mixture-of-experts

    The May 2024 report presents DeepSeek-V2, a mixture-of-experts model with 236 billion total parameters and 21 billion active per token. Multi-head Latent Attention and DeepSeekMoE support efficient inference and economical training.

  3. November 2024

    Fire-Flyer at SC24

    At SC24 in November 2024, the program lists Su with DeepSeek AI on the Fire-Flyer AI-HPC paper. The report describes Fire-Flyer 2 training on 10,000 PCIe A100 GPUs, with performance approximating DGX-A100 at half the cost and 40% less energy.

  4. December 2024

    DeepSeek-V3 technical report

    The December 2024 report presents DeepSeek-V3, a mixture-of-experts model with 671 billion total parameters and 37 billion active for each token. Multi-head Latent Attention and DeepSeekMoE support efficient inference and cost-effective training.

  5. January 2025

    DeepSeek-R1 reasoning

    In January 2025, Su coauthors DeepSeek-R1, which investigates how pure reinforcement learning can develop reasoning without human-labeled reasoning trajectories.

  6. December 2025

    DeepSeek-V3.2 sparse attention

    DeepSeek-V3.2's December 2025 report introduces DeepSeek Sparse Attention, an efficient mechanism that reduces computation in long-context scenarios while preserving model performance.

  7. April 2026

    DeepSeek-V4 million-token context

    In April 2026, DeepSeek-V4 introduces Pro and Flash models that support one-million-token contexts.

  8. As of October 2026

    Coauthoring DeepSeek-V4.1-Flash

    As of October 2026, Su is listed as a coauthor on DeepSeek-AI's DeepSeek-V4.1-Flash report, submitted in September 2026. The report describes global KV cache compression to 890 bytes per token.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  2. The United States employs an estimated 48% of people shaping AIWe estimate where the people shaping AI work today, how universities, companies and research institutes split across countries, and how country shares have changed since the 1960s.
  3. 43% of living people shaping AI work outside their country of birthWe estimate that 43% of living people shaping AI work outside their country of birth (520 of 1,198 people with known birth and employment countries).
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionWe estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.

Sources and credits

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

  1. DeepSeek LLM: Scaling Open-Source Language Models with Longtermism arxiv.org. Checked 21 October 2026.
  2. DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model arxiv.org. Checked 21 October 2026.
  3. Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning arxiv.org. Checked 21 October 2026.
  4. Presentation – SC24 Schedule sc24.conference-program.com. Checked 21 October 2026.
  5. DeepSeek-V3 Technical Report arxiv.org. Checked 21 October 2026.
  6. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org. Checked 21 October 2026.
  7. DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models arxiv.org. Checked 21 October 2026.
  8. DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models alphaxiv.org. Checked 21 October 2026.
  9. DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence arxiv.org. Checked 21 October 2026.
  10. DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression arxiv.org. Checked 21 October 2026.