- Independent researcher, Independent1
Junyang Lin is an independent researcher working on large language models, multimodal systems and agentic learning, with recent work on agentic coding, multimodal modeling, reinforcement-learning methods and efficient attention design. Lin is a co-author of the OpenHands paper (2024), which presents the MIT-licensed open platform for AI software-development agents formerly called OpenDevin, and Lin's GitHub profile pins the OpenHands repository. Lin's homepage lists selected work on the Qwen3 and Qwen3-VL technical reports, Qwen3-Coder, Group Sequence Policy Optimization (GSPO) and gated attention for language models, and Lin co-authored the Qwen Technical Report (2023).123456
At a glance
- Current role
- Independent researcher
- Company
- Independent
- Sector
- Independent
- Fields
- Language models, Multimodal models, Agents
- Work listed
- 3
- Sources cited
- 11
- Profile reviewed
Research
Lin's homepage frames the research around three questions: how models can plan and act rather than only answer, how language and vision can work as one system, and which training methods support stronger post-training. Themes include long-horizon planning, tool use, multimodal representation and reinforcement-learning infrastructure.1
On training methods, Lin co-authored Group Sequence Policy Optimization (GSPO), a sequence-level policy optimization method for reinforcement learning of large models, and "Gated Attention for Large Language Models" (2025), which studies gating in attention for non-linearity, sparsity and avoiding attention sinks.571
Qwen models and agents
Lin is a co-author of the Qwen Technical Report (2023), the Qwen2.5-VL Technical Report (2025) on vision-language models and the Qwen3 Technical Report (2025). The homepage also lists Qwen3-Coder, an agentic coding model the Qwen team announced in July 2025, and the Qwen-Image technical report on image generation among Lin's selected work.68491
Lin co-authored the 2024 OpenHands paper on an open platform for software-development agents. In a March 2026 essay, "From 'Reasoning' Thinking to 'Agentic' Thinking", Lin describes the shift from static reasoning models to agentic systems that think in order to act.310
Work and contributions
OpenHands (formerly OpenDevin) Open-source code
Co-author of "OpenHands: An Open Platform for AI Software Developers as Generalist Agents" (2024), which introduces OpenHands (f.k.a. OpenDevin) and notes its release under the MIT license; Lin's GitHub profile pins the OpenHands repository.3112
Qwen3 Technical Report Published research
Co-author of the Qwen3 Technical Report (2025), which Lin's homepage lists as selected work; also a co-author of the Qwen Technical Report (2023).416
Group Sequence Policy Optimization (GSPO) Published research
Co-author of "Group Sequence Policy Optimization" (2025), which Lin's homepage describes as research on sequence-level policy optimization for reinforcement learning in large model training.51
Timeline
- Co-authored the Qwen Technical Report.6
- Co-authored the OpenHands paper on an open platform for AI software developers.3
- Co-authored the Qwen2.5-VL Technical Report on vision-language models.8
- Co-authored "Gated Attention for Large Language Models".7
- Co-authored the Qwen3 Technical Report.4
- The Qwen team announced Qwen3-Coder, an agentic coding model listed among Lin's selected work.91
- Co-authored "Group Sequence Policy Optimization" (GSPO).5
- Published the essay "From 'Reasoning' Thinking to 'Agentic' Thinking".10
Where they work
- Independent 2 people in the wiki
Related people
- Alec RadfordAlso at Independent
- Jianfeng GaoLanguage models · Multimodal models
- Gabe PereyraAgents · Language models
- Mira MuratiLanguage models · Multimodal models
- Graham NeubigAgents · Language models
- Jason WestonLanguage models · Agents
- TekniumLanguage models · Agents
- Kari BriskiLanguage models · Agents
- Xiaodong HeMultimodal models · Language models
- Yoav ShohamLanguage models · Agents
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
11 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- Junyang Lin — homepage (about and selected work)
- JustinLin610 (Junyang Lin) on GitHub
- OpenHands: An Open Platform for AI Software Developers as Generalist Agents (arXiv 2407.16741)
- Qwen3 Technical Report (arXiv 2505.09388)
- Group Sequence Policy Optimization (arXiv 2507.18071)
- Qwen Technical Report (arXiv 2309.16609)
- Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free (arXiv 2505.06708)
- Qwen2.5-VL Technical Report (arXiv 2502.13923)
- Qwen3-Coder: Agentic Coding in the World
- From 'Reasoning' Thinking to 'Agentic' Thinking
- OpenHands/OpenHands LICENSE (MIT)