Who Is Junyang Lin, Former Qwen Technical Lead?
The former Alibaba engineering lead behind the open-source Qwen models and founder of Pragmatik Labs.
Junyang Lin is an artificial intelligence (AI) researcher and the founder of Pragmatik Labs, who previously served as the technical lead of the Qwen open-source model series at Alibaba.
At Layer3Labs, we build AI agents for small-business workflows, and Qwen models are among the open models a client team can pick to run an agent on.
Lin resigned from Alibaba in early March 2026 and launched Pragmatik Labs in Shanghai with backing from Tencent and major venture capital firms to develop digital and physical agents.
Where Junyang Lin Is Now
Junyang Lin runs Pragmatik Labs, an artificial intelligence laboratory founded in Shanghai after he stepped down as technical lead of Alibaba's Qwen team in March 2026. Lin announced Pragmatik Labs on X on August 12, 2026, as reported by Dealroom and The Next Web. The Next Web describes the lab's goal as next-generation agents across digital and physical worlds.
Lin has not given a public reason for leaving. Reporting from Caixin indicated that Lin sought greater independence for the Qwen team's AI infrastructure and favored tighter integration between pretraining and post-training phases.
Pragmatik Labs is developing digital agents for knowledge work alongside physical agents designed to help robotic systems execute real-world tasks. At its August 12 launch, The Next Web reported that Pragmatik Labs had no released model, product or demo.
His name appears in Western order as Junyang Lin and in Chinese order as Lin Junyang. Both forms refer to the same person.
- Current role: Founder of Pragmatik Labs.
- Organization: Pragmatik Labs, based in Shanghai.
- Known for: Leading Alibaba's Qwen model team from around 2022 until March 2026.
- Education: Undergraduate degree in English, graduate study in foreign linguistics and applied linguistics at Peking University.
- In the news for: Resigning from Alibaba in March 2026, then launching Pragmatik Labs in August 2026 with venture backing from Tencent, Gaorong Ventures, and HSG.
Departure from Alibaba and Tongyi Lab
Lin submitted his resignation from Alibaba on March 3, 2026, creating an unexpected leadership transition for the Qwen open-source model family. TechNode reported the resignation date. Lin announced the move on X, per KrASIA: "me stepping down. bye my beloved qwen." TechNode noted that Lin had led development work on the Qwen model series under Alibaba Cloud Chief Technology Officer (CTO) Zhou Jingren.
KrASIA reported that Alibaba Chief Executive Officer (CEO) Eddie Wu convened an emergency all-hands meeting at Tongyi Lab on March 4, 2026, reassuring staff that Qwen remained a top priority for Alibaba. KrASIA described Lin as the "main driver of Qwen's technical direction" and noted that Hao Zhou, formerly of Google DeepMind, was expected to assume post-training leadership for Qwen.
TechNode reported that the departure coincided with planned organizational adjustments inside Tongyi Lab to reorganize its model development teams. The departure occurred just one day after Alibaba open-sourced the Qwen3.5 small-model series on March 2, 2026.
Academic Background and Early Research
Caixin reported that Lin completed an undergraduate degree majoring in English. He then did graduate study at Peking University, where he specialized in foreign linguistics and applied linguistics.
In 2019, Lin joined Alibaba's DAMO Academy as an algorithm engineer.
Technical Leadership and the Growth of Qwen
Lin directed technical development for the Qwen model family from around 2022 until his departure in 2026. Alibaba established Tongyi Lab around 2022, appointing Lin to head the Qwen large-model project under Alibaba Cloud CTO Zhou Jingren. Alibaba promoted Lin to P10 in 2025, which represents one of the highest individual technical ranks within Alibaba.
Caixin calls Qwen one of the world's most popular open-source AI model series, with more than 1 billion downloads and over 200,000 derivative models.
If you are weighing a current Qwen model, our Qwen 3.8 benchmarks page covers test results, how to use Qwen 3.8 Max covers setup, and Qwen 3.8 alternatives compares other model families.
Pragmatik Labs and Autonomous Agent Architecture
Pragmatik Labs develops artificial intelligence agents designed to handle knowledge workflows and physical robotic tasks. Lin publicly announced Pragmatik Labs on August 12, 2026, establishing the business in Shanghai. According to The Next Web, the name Pragmatik reflects linguistic pragmatics, which Lin described as "being back to where it happens."
The technical roadmap for Pragmatik Labs divides into four main focus areas: digital agents for knowledge work, physical agents that enable robots to execute real-world tasks, frontier foundation research, and iterative products shaped by operational feedback.
Across the workflows we have automated for SMB teams, language models reliably generate drafts but stumble when orchestrating external software tools without strict verification layers. Pragmatik Labs has not yet shown how its agents handle that problem: The Next Web reported no released model, product or demo at the August 12 launch.
Other Chinese AI founders we profile include DeepSeek founder Liang Wenfeng and Moonshot AI founder Yang Zhilin.
All three appear on our AI people to watch list.
Pragmatik Labs Funding and Valuation
Pragmatik Labs closed an initial financing round raising several hundred million dollars at a post-money valuation of about $2 billion, according to reporting by The Information. Reporting published via Dealroom in June 2026 detailed that Tencent contributed $20 million, while venture capital firms Gaorong Ventures and HSG (formerly Sequoia China) co-led the financing, investing approximately $100 million each. The Shanghai Future Industries Fund, also known as the Shanghai Engine Fund, participated as an institutional investor.
Discrepancies exist across financial reports regarding round specifics. While The Information reported Tencent's investment at $20 million, The Next Web described Tencent's stake as an undisclosed amount. Valuation figures also show variations: The Information placed the post-money valuation at approximately $2 billion, while the same Dealroom note also carries a $1.29 billion valuation line. An August 12 summary from Dealroom mistakenly asserted that Pragmatik Labs raised approximately $2 billion, conflating total valuation with funding raised.
No named financial outlet has published an estimated net worth or executive compensation figure for Junyang Lin as of October 2026.
Frequently Asked Questions
- Alibaba owns the Qwen artificial intelligence model series. Tongyi Lab developed the models under Alibaba Cloud leadership, with Junyang Lin serving as technical lead until early March 2026.
- Junyang Lin has not publicly stated his reasons for leaving Qwen. Reporting from Caixin indicated internal disagreements regarding the integration of pretraining and post-training workflows, as well as Lin's push for greater independence for Qwen's artificial intelligence infrastructure. TechNode reported that the departure coincided with planned organizational adjustments inside Tongyi Lab to reorganize its model development teams.
- Qwen is an open-source family of AI models from Alibaba that developers download and build on. According to Caixin, the open-source Qwen model family has surpassed 1 billion downloads and powered more than 200,000 derivative models worldwide.
- Following Lin's resignation in early March 2026, Alibaba Chief Executive Officer Eddie Wu held an emergency all-hands meeting at Tongyi Lab on March 4, 2026, affirming that Qwen remained a top corporate priority. KrASIA reported that Hao Zhou, formerly of Google DeepMind, was expected to assume post-training leadership for Qwen.
- Yes, Alibaba created the Qwen model family through Tongyi Lab. Junyang Lin led development work on the model series under Alibaba Cloud Chief Technology Officer Zhou Jingren until Lin submitted his resignation on March 3, 2026.
Evaluate Foundation Models for Enterprise Workflows
Speak with our implementation team about benchmarking open-source models, agent architectures, and autonomous execution pipelines for your business operations.
Book a Consultation