PeopleofSuperintelligence

Researcher

Liyue Zhang

Liyue Zhang is an artificial intelligence researcher at DeepSeek whose name appears on the company's open language models and on the computing systems that train them. Zhang coauthored DeepSeek LLM in January 2024, then the Mixture-of-Experts models DeepSeek-V2 in May 2024 and DeepSeek-V3 in December 2024. As a member of the DeepSeek-AI Team in Hangzhou, they also helped create DeepSeek-R1, a reasoning model trained with pure reinforcement learning that Nature published in September 2025, along with the Prover models for theorem proving in Lean 4. Their systems papers include Fire-Flyer AI-HPC, a 2024 design for a cluster of 10,000 PCIe A100 GPUs, and Insights into DeepSeek-V3, which appeared in the ISCA 2025 proceedings. In 2026, they coauthored DeepSeek-V4, DeepSeek-V4.1-Flash and the DSec sandbox platform for agentic training.

Now at
DeepSeek
Works
4
Citations
46
h-index
1

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. January 2024

    DeepSeek LLM and scaling laws

    In January 2024, Zhang coauthors DeepSeek LLM, a study of scaling laws that trains open models with 7B and 67B parameters. The project builds a pre-training dataset of 2 trillion tokens and creates DeepSeek Chat models through supervised fine-tuning and Direct Preference Optimization.

  2. May 2024

    The DeepSeek-V2 model

    DeepSeek-V2, which Zhang coauthors in May 2024, is a Mixture-of-Experts language model with 236B total parameters, 21B of them active for each token. Its Multi-head Latent Attention compresses the key-value cache, which shrinks by 93.3% compared with DeepSeek 67B.

  3. August 2024

    Prover-V1.5 and Fire-Flyer

    In August 2024, Zhang coauthors DeepSeek-Prover-V1.5, an open-source model for theorem proving in Lean 4 that learns from proof assistant feedback through reinforcement learning. They also coauthor the Fire-Flyer AI-HPC paper, accepted at SC'24, which describes a cluster of 10,000 PCIe A100 GPUs that approaches DGX-A100 performance at half the cost.

  4. December 2024

    The DeepSeek-V3 model

    In December 2024, Zhang coauthors the DeepSeek-V3 technical report. The Mixture-of-Experts model has 671B total parameters with 37B active for each token, is pre-trained on 14.8 trillion tokens and needs 2.788M H800 GPU hours for its full training.

  5. January 2025–September 2025

    DeepSeek-R1 in Nature

    In January 2025, they coauthor DeepSeek-R1, which uses pure reinforcement learning to build reasoning abilities such as self-reflection and verification in large language models. Nature publishes the work in September 2025, with Zhang credited as part of the DeepSeek-AI Team in Hangzhou.

  6. April 2025–May 2025

    Prover-V2 and an ISCA paper

    In April 2025, Zhang coauthors DeepSeek-Prover-V2, whose 671B version reaches an 88.9% pass ratio on the MiniF2F-test for formal theorem proving in Lean 4. In May 2025, they publish Insights into DeepSeek-V3 with Chenggang Zhao, Wenfeng Liang and other colleagues, an ISCA '25 industry-track paper on how training on 2,048 NVIDIA H800 GPUs shaped the model's hardware-aware design.

  7. December 2025

    The DeepSeek-V3.2 model

    With DeepSeek-V3.2 in December 2025, Zhang and colleagues introduce DeepSeek Sparse Attention, which reduces computational complexity in long-context scenarios. A high-compute variant, DeepSeek-V3.2-Speciale, reaches gold-medal performance at the 2025 International Mathematical Olympiad and the International Olympiad in Informatics.

  8. As of October 2026

    Researcher at DeepSeek

    Zhang works at DeepSeek as an artificial intelligence researcher. In April 2026, they coauthor the preview of DeepSeek-V4, whose Pro model has 1.6T parameters and a one-million-token context. September 2026 brings DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts model, and DSec, a sandbox platform for agentic training, both with Zhang as a coauthor.

Most-cited works

  1. Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures2025 · International Symposium on Computer Architecture · 42 citationsWith Chengqi Deng, Damai Dai, Huazuo Gao and Shangyan Zhou
  2. DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence2026 · 4 citationsWith Bingxuan Wang, Chengqi Deng, Damai Dai, Daya Guo, Erhang Li and Fangyun Lin
  3. DualPath: Breaking the Storage Bandwidth Bottleneck in Agentic LLM Inference2026 · 1 citationsWith Shangyan Zhou
  4. DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression2026 · 0 citationsWith Chengqi Deng, Damai Dai, Erhang Li, Fangyun Lin, Fucong Dai and Guanting Chen

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Liyue Zhang. 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 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 Liyue Zhang 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. [2405.04434] DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model arxiv.org. Checked 20 October 2026.
  3. [2408.08152] DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search arxiv.org. Checked 20 October 2026.
  4. [2408.14158] Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning arxiv.org. Checked 20 October 2026.
  5. [2412.19437] DeepSeek-V3 Technical Report arxiv.org. Checked 20 October 2026.
  6. [2501.12948] DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org. Checked 20 October 2026.
  7. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning | Nature nature.com. Checked 20 October 2026.
  8. [2504.21801] DeepSeek-Prover-V2: Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decomposition arxiv.org. Checked 20 October 2026.
  9. [2505.09343] Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures arxiv.org. Checked 20 October 2026.
  10. Paper page - Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures huggingface.co. Checked 20 October 2026.
  11. [2512.02556] DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models arxiv.org. Checked 20 October 2026.
  12. Liyue Zhang - Wikidata wikidata.org. Checked 20 October 2026.
  13. [2606.19348] DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence arxiv.org. Checked 20 October 2026.
  14. [2609.19969] DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression arxiv.org. Checked 20 October 2026.
  15. [2609.22978] DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale arxiv.org. Checked 20 October 2026.