Junyang Lin

HSS-Synth generates 237k high-quality instruction samples that not only outperform existing baselines but also redefine data synthesis for the humanities and social sciences.

Current AI agents only manage to complete 20.6% of complex real-world tasks, revealing a stark gap in their capabilities compared to human users.

OmniAgent not only outperforms larger models but also scales performance with reasoning turns, revolutionizing how we approach video understanding.

Qwen-RobotManip achieves a 20% relative improvement over the previous state-of-the-art in robotic manipulation, showcasing unprecedented generalization capabilities from diverse, open-source datasets.

A single visual tokenizer in UniAR bridges the gap between understanding and generation, achieving state-of-the-art performance in image generation and editing.

Qwen-RobotNav redefines navigation by allowing real-time reconfiguration of strategies, achieving unprecedented flexibility and performance across diverse tasks.

MTP acceptance rates can be dramatically improved by addressing entropy fluctuations, leading to up to 1.8x faster RL training.

LLMs can switch between reasoning and factual answering on the fly, without retraining, simply by conditioning on specific token prefixes.

Multimodal models are often blind at birth: a new "Visual Attention Score" reveals they struggle to focus on visual inputs during cold-start, but a simple attention-guided fix can boost performance by 7%.

An 80B model that runs like a 3B? Qwen3-Coder-Next shows you can get competitive coding agent performance with a fraction of the active parameters, thanks to smart training.

LLM benchmark accuracy jumps 10% when evaluated on a cleaned-up version of Humanity's Last Exam, highlighting the significant impact of dataset noise on performance metrics.

ToolRMs drastically improve tool-use accuracy in LLMs, outperforming existing models by up to 17.94%, while also reducing output token usage by over 66% through efficient inference-time scaling.