Daily Papers of 2025-12-25

  1. TurboDiffusion: Accelerating Video Diffusion Models by 100-200 Times 88 upvotes, #1 of 2025-12-25
  2. Learning to Reason in 4D: Dynamic Spatial Understanding for Vision Language Models 48 upvotes, #2 of 2025-12-25
  3. DreaMontage: Arbitrary Frame-Guided One-Shot Video Generation 32 upvotes, #3 of 2025-12-25
  4. Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning 28 upvotes, #4 of 2025-12-25
  5. NVIDIA Nemotron 3: Efficient and Open Intelligence 27 upvotes, #5 of 2025-12-25
  6. Beyond Memorization: A Multi-Modal Ordinal Regression Benchmark to Expose Popularity Bias in Vision-Language Models 26 upvotes, #6 of 2025-12-25
  7. T2AV-Compass: Towards Unified Evaluation for Text-to-Audio-Video Generation 24 upvotes, #7 of 2025-12-25
  8. HiStream: Efficient High-Resolution Video Generation via Redundancy-Eliminated Streaming 20 upvotes, #8 of 2025-12-25
  9. DramaBench: A Six-Dimensional Evaluation Framework for Drama Script Continuation 16 upvotes, #9 of 2025-12-25
  10. TokSuite: Measuring the Impact of Tokenizer Choice on Language Model Behavior 16 upvotes, #9 of 2025-12-25
  11. Learning from Next-Frame Prediction: Autoregressive Video Modeling Encodes Effective Representations 12 upvotes, #11 of 2025-12-25
  12. From Word to World: Can Large Language Models be Implicit Text-based World Models? 11 upvotes, #12 of 2025-12-25
  13. SWE-EVO: Benchmarking Coding Agents in Long-Horizon Software Evolution Scenarios 9 upvotes, #13 of 2025-12-25
  14. Streaming Video Instruction Tuning 9 upvotes, #13 of 2025-12-25
  15. Multi-hop Reasoning via Early Knowledge Alignment 6 upvotes, #15 of 2025-12-25
  16. LLM Swiss Round: Aggregating Multi-Benchmark Performance via Competitive Swiss-System Dynamics 2 upvotes, #16 of 2025-12-25
  17. PhononBench:A Large-Scale Phonon-Based Benchmark for Dynamical Stability in Crystal Generation 1 upvotes, #17 of 2025-12-25

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.