Daily Papers of 2026-10-06

  1. ALoDLM: Adaptively Looped Diffusion Language Models 29 upvotes, #1 of 2026-10-06
  2. Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation 11 upvotes, #2 of 2026-10-06
  3. CANOPY: Adaptive-Granularity Evidence Compression for Multimodal RAG 10 upvotes, #3 of 2026-10-06
  4. ASCENT: Online Test-Time Training of Long-Horizon Agents via Self-Distillation of Verified Experience 10 upvotes, #3 of 2026-10-06
  5. Rethinking Long-Video Efficiency: A Joint Allocation Perspective on Frames, Pixels, and Front-End Latency 7 upvotes, #5 of 2026-10-06
  6. OSWorld-Pro: Process-based Evaluation for Computer Use Agents 6 upvotes, #6 of 2026-10-06
  7. Towards Looped Models Done Right, Part II: Rethinking at Fixed Points 5 upvotes, #7 of 2026-10-06
  8. Prism: Dynamic Sparse Attention for Native 2K Joint Video-Audio Generation Model Training 3 upvotes, #8 of 2026-10-06
  9. PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins 2 upvotes, #9 of 2026-10-06
  10. PaLoRA: Paced Low-Rank Adaptation for Continual Learning 2 upvotes, #9 of 2026-10-06
  11. Dynamic Harness Search: Building Multi-Agent Systems Per-Query via Prediction 0 upvotes, #11 of 2026-10-06
  12. When to Switch: Reliable Action-Chunk Extension for Vision-Language-Action Models 0 upvotes, #11 of 2026-10-06
  13. Certification of Real Images through Calibrated Content Authentication 0 upvotes, #11 of 2026-10-06
  14. Noise Out, Bias In: Targeted Bias Injection in Diffusion Language Models via Closed-Loop Activation Steering 0 upvotes, #11 of 2026-10-06
  15. RealtimeWAM: One-Step Asynchronous World Action Models 0 upvotes, #11 of 2026-10-06
  16. InterMimicGen: Scaling Humanoid Loco-Manipulation through Self-Evolving Motion Imitation 0 upvotes, #11 of 2026-10-06
  17. Base Models Can Reason By Taking a Cue From Training Data 0 upvotes, #11 of 2026-10-06

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.