PeopleofSuperintelligence

Researcher

Haoran Wei

Haoran Wei is an AI researcher whose work connects computer vision, document understanding and language models. He coauthored Vary and General OCR Theory, which explore vision vocabulary expansion and end-to-end document recognition. He coauthored Qwen2, Qwen2.5 and Qwen3 reports, then coauthored DeepSeek-OCR systems for document context compression. As of October 2026, he is a researcher at DeepSeek specializing in computer vision and multimodal language models.

Timeline

  1. 2023

    Vary's vision vocabulary

    In 2023, he coauthors Vary, which expands a large vision-language model's visual vocabulary for fine-grained perception, including document OCR.

  2. July 2024

    Qwen2 language and multimodal models

    In July 2024, he coauthors Qwen2, a series of large language and multimodal models ranging from 0.5 to 72 billion parameters.

  3. September 2024

    General OCR Theory and GOT

    In September 2024, he coauthors General OCR Theory and its GOT model, which handles text, formulas, tables, charts, sheet music and geometric shapes through an end-to-end OCR system.

  4. December 2024

    Qwen2.5 trained on 18 trillion tokens

    In December 2024, he coauthors Qwen2.5, a large language model series trained on datasets scaled from 7 trillion to 18 trillion tokens.

  5. May 2025

    Qwen3's dense and MoE models

    In May 2025, he coauthors Qwen3, a model family developed by Alibaba Cloud's Qwen Team, with dense and mixture-of-experts models from 0.6 to 235 billion parameters.

  6. October 2025

    Optical compression with DeepSeek-OCR

    In October 2025, he coauthors DeepSeek-OCR, which compresses long contexts through optical 2D mapping and reports 97% OCR precision when text tokens are within 10 times vision tokens.

  7. As of October 2026

    Researcher at DeepSeek

    In January 2026, he coauthors DeepSeek-OCR 2, whose encoder uses causal reasoning to reorder visual tokens by image semantics. As of October 2026, he is a researcher at DeepSeek specializing in computer vision and multimodal language models.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Sources and credits

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

  1. [2312.06109] Vary: Scaling up the Vision Vocabulary for Large Vision-Language Models arxiv.org. Checked 19 October 2026.
  2. [2407.10671] Qwen2 Technical Report arxiv.org. Checked 19 October 2026.
  3. [2409.01704] General OCR Theory: Towards OCR-2.0 via a Unified End-to-end Model arxiv.org. Checked 19 October 2026.
  4. [2412.15115] Qwen2.5 Technical Report arxiv.org. Checked 19 October 2026.
  5. [2505.09388] Qwen3 Technical Report arxiv.org. Checked 19 October 2026.
  6. GitHub - QwenLM/Qwen3: Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud. github.com. Checked 19 October 2026.
  7. [2510.18234] DeepSeek-OCR: Contexts Optical Compression arxiv.org. Checked 19 October 2026.
  8. [2601.20552] DeepSeek-OCR 2: Visual Causal Flow arxiv.org. Checked 19 October 2026.
  9. Haoran Wei | alphaXiv alphaxiv.org. Checked 19 October 2026.