mlsanityData to 7 Oct 2026

Visual token pruning in multimodal models

520 papers4 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: visual, token, tokens, pruning, visual tokens

Key facts. Visual token pruning in multimodal models: 520 papers on arXiv since 2021 (data through 7 Oct 2026). 243 of its 520 mapped papers (47%) cite “An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models” (2024, arXiv 2403.06764). 182 of its 520 mapped papers (35%) cite “LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models” (2024, arXiv 2403.15388). 22 new papers 24 Sep–7 Oct.

What’s new

24 Sep – 7 Oct 2026

22 new papers, 24 Sep – 7 Oct; 4 this week.

  1. This week · 4
  2. Adaptive Visual Token Reduction for Accelerated Image Understanding
  3. DIPrune: Task-Aware Token Pruning with Dual Importance for Efficient Multimodal Language Models
  4. VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models
  5. VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMs
  6. Earlier · 18
  7. When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLA
  8. MWOP: Modality-aware Width-wise Operation Pruning for Efficient MLLMs
  9. CoVisco: Codec-Native Vision Encoder with Native Token Compression for Unified Image-Video Understanding also fits: Token pruning and merging for ViTs
  10. When Masking Helps or Hurts Robustness in Compressed CLIP: A Pre-Deployment Diagnostic
  11. Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models
  12. TReVS: Integrating Textual Relevance and Visual Saliency for Efficient Vision-Language Model Token Pruning
All 22
  1. GleanVID: Complementary Token Selection for Efficient Video Large Language Models
  2. Representation Dynamics Reveal Semantic Saliency and Similarity for Visual Token Pruning in MLLMs
  3. Rethinking Visual Token Compression for Video Large Language Models: A Simple Yet Strong Baseline
  4. Beyond Selection: Token Parameterization for Extreme Visual Token Compression
  5. Still There, No Longer Seen: Exposing Compression-Induced Risk in Large Vision-Language Models
  6. SPIDER: Multi-Layer Semantic Token Pruning and Adaptive Sub-Layer Skipping in Multimodal Large Language Models
  7. P4Q: Co-designing Token Pruning and Quantization for Vision-Language Model Acceleration
  8. When Text Matters: Design Principles for Visual Token Pruning in Vision-Language Model
  9. ACPruner: Visual Token Pruning as Biased Attention Coverage Maximization in LVLMs
  10. MiCo: Mutual Information Coverage Optimization through Semantic Erasure Modeling for Efficient MLLM Inference
  11. CoViST: Visual Token Compression via Composable States
  12. Fewer Tokens, More Self-Teaching: On-Policy Self-Distillation for Extreme Visual Token Reduction

Foundations

Most cited by papers here

  1. An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models2024cited by 243 here
  2. LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models2024cited by 182 here
  3. DivPrune: Diversity-based Visual Token Pruning for Large Multimodal Models2025cited by 112 here
  4. Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters More2025cited by 95 here
  5. Beyond Text-Visual Attention: Exploiting Visual Cues for Effective Token Pruning in VLMs2024cited by 89 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022Token Merging: Your ViT But Fastercited by 212 here
  2. 2024LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Modelscited by 182 here
  3. 2024LOOK-M: Look-Once Optimization in KV Cache for Efficient Multimodal Long-Context Inferencecited by 27 here
  4. 2024DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Modelscited by 39 here
  5. 2025TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Modelcited by 42 here
  6. 2025Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMscited by 62 here
18 more it builds on, not in this chain
  1. 2023Token Merging for Fast Stable Diffusioncited by 28 here
  2. 2023H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Modelscited by 38 here
  3. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 31 here
  4. 2024An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Modelscited by 243 here
  5. 2024SnapKV: LLM Knows What You are Looking for Before Generationcited by 38 here
  6. 2024HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Modelscited by 60 here
  7. 2024Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Modelscited by 67 here
  8. 2024Multi-Stage Vision Token Dropping: Towards Efficient Multimodal Large Language Modelcited by 39 here
  9. 2024ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Modelscited by 49 here
  10. 2024Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsificationcited by 28 here
  11. 2024Beyond Text-Visual Attention: Exploiting Visual Cues for Effective Token Pruning in VLMscited by 89 here
  12. 2024PruneVid: Visual Token Pruning for Efficient Video Large Language Modelscited by 31 here
  13. 2024FrameFusion: Combining Similarity and Importance for Video Token Reduction on Large Vision Language Modelscited by 26 here
  14. 2025Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters Morecited by 95 here
  15. 2025Token Pruning in Multimodal Large Language Models: Are We Solving the Right Problem?cited by 35 here
  16. 2025DivPrune: Diversity-based Visual Token Pruning for Large Multimodal Modelscited by 112 here
  17. 2025PACT: Pruning and Clustering-Based Token Reduction for Faster Visual Language Modelscited by 32 here
  18. 2025Don't Just Chase "Highlighted Tokens" in MLLMs: Revisiting Visual Holistic Context Retentioncited by 28 here

The 24 most cited of 520.

Browse

30 of 520, ranked 6–35

  1. 6Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Models202467
  2. 7Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs202562
  3. 8HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models202460
  4. 9ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models202449
  5. 10TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model202542
  6. 11Multi-Stage Vision Token Dropping: Towards Efficient Multimodal Large Language Model202439
  7. 12DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models202439
  8. 13Token Pruning in Multimodal Large Language Models: Are We Solving the Right Problem?202535
  9. 14PACT: Pruning and Clustering-Based Token Reduction for Faster Visual Language Models202532
  10. 15PruneVid: Visual Token Pruning for Efficient Video Large Language Models202431
All 30
  1. 16Don't Just Chase "Highlighted Tokens" in MLLMs: Revisiting Visual Holistic Context Retention202528
  2. 17Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification202428
  3. 18Token Merging for Fast Stable Diffusion202328
  4. 19FrameFusion: Combining Similarity and Importance for Video Token Reduction on Large Vision Language Models202426
  5. 20HoliTom: Holistic Token Merging for Fast Video Large Language Models202525
  6. 21DeCo: Decoupling Token Compression from Semantic Abstraction in Multimodal Large Language Models202425
  7. 22Filter, Correlate, Compress: Training-Free Token Reduction for MLLM Acceleration202425
  8. 23What Kind of Visual Tokens Do We Need? Training-free Visual Token Pruning for Multi-modal Large Language Models from the Perspective of Graph202524
  9. 24MADTP: Multimodal Alignment-Guided Dynamic Token Pruning for Accelerating Vision-Language Transformer202424
  10. 25VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models202523
  11. 26Feather the Throttle: Revisiting Visual Token Pruning for Vision-Language Model Acceleration202423
  12. 27Efficient Large Multi-modal Models via Visual Context Compression202422
  13. 28AIM: Adaptive Inference of Multi-Modal LLMs via Token Merging and Pruning202419
  14. 29Video Compression Commander: Plug-and-Play Inference Acceleration for Video Large Language Models202517
  15. 30EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models202516
  16. 31ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsification202416
  17. 32VL-Cache: Sparsity and Modality-Aware KV Cache Compression for Vision-Language Model Inference Acceleration202416
  18. 33Accelerating Multimodal Large Language Models by Searching Optimal Vision Token Reduction202416
  19. 34Streaming Video Question-Answering with In-context Video KV-Cache Retrieval202515
  20. 35Global Compression Commander: Plug-and-Play Inference Acceleration for High-Resolution Large Vision-Language Models202514