Daily Papers of 2026-08-21

  1. EnvHarness: Awakening Static Worlds for Agent Learning 265 upvotes, #1 of 2026-08-21
  2. 4DAnyone: Create Anyone in 4D from a Casual Monocular Video 80 upvotes, #2 of 2026-08-21
  3. SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science? 64 upvotes, #3 of 2026-08-21
  4. WithEveryone: Unified Planning and Identity Grounding for Group Image Generation 42 upvotes, #4 of 2026-08-21
  5. MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use 33 upvotes, #5 of 2026-08-21
  6. SkillEvo: Self-Renewing Evolution Gradients from Multi-Turn Interaction Feedback 31 upvotes, #6 of 2026-08-21
  7. ForgeWM: Progressive Causal Training for Few-Step Action-Conditioned Video World Models 24 upvotes, #7 of 2026-08-21
  8. FlowEvo: Self-Evolving Agents through the Co-Evolution of Workflows and Executable Skills 20 upvotes, #8 of 2026-08-21
  9. Repo0: Design-Driven Zero-to-All Code Generation 20 upvotes, #8 of 2026-08-21
  10. FlashPrefill V2: Block-Sparse Prefill Attention for Long-Context LLM Serving 19 upvotes, #10 of 2026-08-21
  11. EXIMO: VLM Guided Exploration of VLA Policies 16 upvotes, #11 of 2026-08-21
  12. The Embedder's Dilemma: LLMs Are Better, but at What Cost? 15 upvotes, #12 of 2026-08-21
  13. τ_0-VLA: a Hierarchical Robot Foundation Model with World-Model-Guided Test-Time Computation 15 upvotes, #12 of 2026-08-21
  14. Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See 15 upvotes, #12 of 2026-08-21
  15. Hierarchical Self-Improvement: A Framework for Task-Specific Evolvable Agent Harnesses 12 upvotes, #15 of 2026-08-21
  16. Inject, Align, Recover: Staged Post-Training for Retrieval-Free Document Knowledge Internalization 12 upvotes, #15 of 2026-08-21
  17. Towards Quantifying Benchmark Optimization in ASR Models 11 upvotes, #17 of 2026-08-21
  18. NARU: A Benchmark for NARrative Evolution and Cultural Nuance Understanding in Japanese Extreme Long Video 9 upvotes, #18 of 2026-08-21
  19. TinyCast: Probabilistic Zero-Shot Forecasting with Computed Periodicity 9 upvotes, #18 of 2026-08-21
  20. Chain-of-Experience for Continual LLM Improvement 9 upvotes, #18 of 2026-08-21
  21. PolicyGuide: From Guarding One Action to Guiding the Whole Workflow for Policy-Compliant LLM Agents 9 upvotes, #18 of 2026-08-21
  22. QuoteBench: How Matched Scores Can Hide Command-Path Failures 8 upvotes, #22 of 2026-08-21
  23. GOAG: Generative and Object-Agnostic Grasp Planner for Dexterous Robotic Manipulation 8 upvotes, #22 of 2026-08-21
  24. CoToGrasp: Contact-Topology-Conditioned Dexterous Grasp Synthesis via Canonical Workspace Learning 7 upvotes, #24 of 2026-08-21
  25. Listening Forward: Next Patch Embedding Prediction Enables Scalable Audio Learners 6 upvotes, #25 of 2026-08-21

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.