PeopleofSuperintelligence

Researcher

Jianhong Tu

Jianhong Tu is an AI researcher at Alibaba Group who co-authored the technical reports for Qwen, Qwen2, Qwen2.5 and Qwen3, the company's family of large language models. Before Qwen, they co-authored RPT, a 2020 pre-trained Transformer for data preparation, and in 2022 they were first author of a paper on contextual expressive text-to-speech with Alibaba DAMO Academy and Renmin University of China. Their recent work centers on AI agents: in 2026 they were corresponding author of the DeepPlanning benchmark and co-authored WebWorld and Qwen-AgentWorld, world models that simulate environments for agents. They now work in Alibaba Group's Alibaba Token Hub, where they co-authored MetaExp in October 2026, a framework that trains agents to learn from experience.

Now at
Alibaba

Timeline

  1. December 2020

    RPT for data preparation

    Tu co-authors RPT, a Relational Pre-trained Transformer, with Nan Tang, Ju Fan, Sam Madden and others in December 2020. The model pairs a BERT-like encoder with a GPT-like decoder and supports data preparation tasks such as data cleaning, auto-completion and schema matching.

  2. November 2022

    Contextual expressive text-to-speech

    In November 2022, they are first author of Contextual Expressive Text-to-Speech, written with colleagues at Alibaba DAMO Academy and Renmin University of China. The paper introduces Contextual TTS, a task in which written context guides expressive speech synthesis.

  3. September 2023

    The first Qwen technical report

    In September 2023, they co-author the Qwen Technical Report, which introduces the first Qwen language model series. It covers base Qwen models, Qwen-Chat models tuned with human alignment techniques, and the Code-Qwen and Math-Qwen-Chat variants.

  4. July 2024–September 2024

    Qwen2 and Qwen2.5-Math

    The Qwen2 Technical Report, co-authored by Tu in July 2024, describes dense and Mixture-of-Experts models from 0.5 to 72 billion parameters with openly released weights. In September 2024, they co-author the Qwen2.5-Math report on math-specific models built with a self-improvement pipeline.

  5. December 2024–January 2025

    Qwen2.5 and million-token context

    In December 2024, they co-author the Qwen2.5 Technical Report, which scales pre-training data from 7 trillion to 18 trillion tokens. Their Qwen2.5-1M report in January 2025 extends the models' context length to 1 million tokens.

  6. May 2025

    The Qwen3 technical report

    Tu co-authors the Qwen3 Technical Report in May 2025. Qwen3 spans dense and Mixture-of-Experts models from 0.6 to 235 billion parameters and combines thinking and non-thinking modes in one framework.

  7. January 2026–June 2026

    DeepPlanning and agent world models

    Working in the Qwen Team at Alibaba Group, they are corresponding author of DeepPlanning in January 2026, a benchmark of multi-day travel planning and multi-product shopping tasks for long-horizon agents. In February 2026, they co-author WebWorld, an open-web simulator trained on more than 1 million interactions. June 2026 brings Qwen-AgentWorld, language world models that simulate agentic environments across 7 domains, with Tu among its authors.

  8. As of October 2026

    Agent research at Alibaba Token Hub

    In September 2026, from Alibaba Token Foundry, Tu co-authors VHD-Play, a pipeline that produces 3,300 agentic training environments at a few cents each. As of October 2026, they work in Alibaba Group's Alibaba Token Hub and co-author MetaExp, a framework that trains agents to turn past interactions into reusable experience across verifiable task families.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Alibaba. 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.
  2. Universities (31%) and startups and other companies (25%) employ the largest shares of 2020–2026 entrantsIncludes Alibaba. Among 819 people whose first recorded role started in 2020–2026 and who have a known current organization, 60% work in commercial sectors (25% at startups and other companies, 23% at frontier labs and 13% at big tech, with 9 in two), and 31% work at universities, compared with 66% for earlier entrants.
  3. The top 10 organizations employed 22% of people shaping AI in 2025: 51% of company researchers and 21% of university researchersIncludes Alibaba. We estimate the share of people shaping AI who worked at the top 10 organizations each year from 1955 to 2025, comparing concentration across companies, universities, and institutes and labs.
  4. 16% of AI career moves in the 2020s went from a university to a companyIncludes Alibaba. 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.

Colleagues

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

Sources and credits

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

  1. [2012.02469] RPT: Relational Pre-trained Transformer Is Almost All You Need towards Democratizing Data Preparation arxiv.org. Checked 20 October 2026.
  2. Contextual Expressive Text-to-Speech arxiv.org. Checked 20 October 2026.
  3. [2309.16609] Qwen Technical Report arxiv.org. Checked 20 October 2026.
  4. [2407.10671] Qwen2 Technical Report arxiv.org. Checked 20 October 2026.
  5. [2409.12122] Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement arxiv.org. Checked 20 October 2026.
  6. [2412.15115] Qwen2.5 Technical Report arxiv.org. Checked 20 October 2026.
  7. [2501.15383] Qwen2.5-1M Technical Report arxiv.org. Checked 20 October 2026.
  8. [2505.09388] Qwen3 Technical Report arxiv.org. Checked 20 October 2026.
  9. Paper page - Qwen3 Technical Report huggingface.co. Checked 20 October 2026.
  10. DeepPlanning: Benchmarking Long-Horizon Agentic Planning with Verifiable Constraints arxiv.org. Checked 20 October 2026.
  11. Paper page - DeepPlanning: Benchmarking Long-Horizon Agentic Planning with Verifiable Constraints huggingface.co. Checked 20 October 2026.
  12. WebWorld: A Large-Scale World Model for Web Agent Training arxiv.org. Checked 20 October 2026.
  13. GitHub - QwenLM/WebWorld: WebWorld is a large-scale web world model that helps train web agents in a simulated browser, avoiding the latency and safety issues of the real web. github.com. Checked 20 October 2026.
  14. [2606.24597] Qwen-AgentWorld: Language World Models for General Agents arxiv.org. Checked 20 October 2026.
  15. Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanisms arxiv.org. Checked 20 October 2026.
  16. MetaExp: Learning to Self-Improve through Experience via Verifiable Cross-Task Environment Scaling arxiv.org. Checked 20 October 2026.