Open R1: Update #2
…Thanks! If you feel interested, can be reached out by chenyang.zhao@sglang.ai Would also be nice to include the performance of Qwen2.5-Math-Instruct and Qwen2.5-7B-Instruct…
Search
Tracked topic
Qwen3
Qwen3 is an AI model family developed by Alibaba, released as a set of large language models for natural-language tasks.
Live now
Qwen3.8-Flash-Next: Time to Update Those Benchmarks
Paper page - Recursive Synthesis for Long-Horizon Terminal Tasks
…To demonstrate training utility, we collect rejection-sampled Qwen3.5 trajectories on the synthesized tasks and use them for supervised fine-tuning . Fine-tuning on these trajectories improves Qwen3.5-27B and…
Paper page - ReVision: Scaling Computer-Use Agents via Temporal Visual Redundancy Reduction
…Generated by Qwen/Qwen2.5-Coder-32B-Instruct Computer-use agents (CUAs) rely on visual observations of graphical user interfaces, where each screenshot is encoded into a large number of visual tokens…
Paper page - DOT-MoE: Differentiable Optimal Transport for MoEfication
…Udbhav Bamba , , , , Abstract DOT-MoE formulates dense layer decomposition as a differentiable optimal transport problem, enabling efficient training of sparse MoE models with improved performance retention. Generated by Qwen/Qwen2.5-Coder…
Paper page - Sumi: Open Uniform Diffusion Language Model from Scratch
…model pretrained from scratch demonstrates competitive performance on knowledge and reasoning tasks while highlighting differences in commonsense reasoning compared to autoregressive models. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Diffusion models…
Paper page - Boosting Omni-Modal Language Models: Staged Post-Training with Visually Debiased Evaluation
…After SFT on self-distilled data, the 3B model reaches performance comparable to, and in aggregate slightly above, Qwen3-Omni-30B-A3B-Instruct without using a stronger omni-modal teacher. These results…
Paper page - Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE
…rescaling factors and utilizing a bifocal attention mechanism that maintains high performance across varying sequence lengths. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Modern LLMs are increasingly deployed in long-context…
Paper page - Online Skill Learning for Web Agents via State-Grounded Dynamic Retrieval
…on current webpage state rather than fixed task-level strategies, improving automation performance across multiple domains. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Language agents increasingly rely on reusable skills to…
Paper page - DanceOPD: On-Policy Generative Field Distillation
…Generated by Qwen/Qwen2.5-Coder-32B-Instruct Modern image generation demands a single model that unifies diverse capabilities, including text-to-image (T2I), local editing , and global editing . However, these capabilities…
Paper page - Benchmarking Visual State Tracking in Multimodal Video Understanding
…state tracking in videos, performing poorly even when human-level capabilities are required, and existing agentic approaches do not effectively address these limitations. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Understanding…