# When Do Fewer Visual Tokens Accelerate Multimodal Inference? A Break-Even Study Across Decision Locations and Hardware

arXiv: https://arxiv.org/abs/2608.03649 · HTML: https://mlsanity.com/paper/2608.03649/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 11 of its 12 references that are in the map.
- First posted on arXiv 4 Aug 2026.

## Builds on (most cited first, 11 of 11)

- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [Towards VQA Models That Can Read](https://mlsanity.com/paper/1904.08920/) (2019)
- [ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning](https://mlsanity.com/paper/2203.10244/) (2022)
- [LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models](https://mlsanity.com/paper/2403.15388/) (2024)
- [SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference](https://mlsanity.com/paper/2410.04417/) (2024)
- [TokenPacker: Efficient Visual Projector for Multimodal LLM](https://mlsanity.com/paper/2407.02392/) (2024)
- [Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Models](https://mlsanity.com/paper/2409.10197/) (2024)
- [AdaptVision: Efficient Vision-Language Models via Adaptive Visual Acquisition](https://mlsanity.com/paper/2512.03794/) (2025)
- [ResAdapt: Adaptive Resolution for Efficient Multimodal Reasoning](https://mlsanity.com/paper/2603.28610/) (2026)
- [CARES: Context-Aware Resolution Selector for VLMs](https://mlsanity.com/paper/2510.19496/) (2025)

## Abstract (arXiv)

Fewer visual tokens do not guarantee lower end-to-end latency. We evaluate break-even with a reproducible protocol that accounts for decision overhead, shared work, and the operators each policy can avoid. A stage-level decomposition reconciles these components with measured end-to-end latency. In a 30-example pilot, the two tested autoregressive probes remain slower than Full despite state reuse. A lightweight post-vision predictor yields paired confidence intervals below zero on RTX 3090 and A100 and remains significant after a conservative all-pairs Holm correction. A pre-vision image-size rule also yields intervals below zero on both GPUs, although neither comparison remains significant after the same correction. Pre-vision routing has a structural opportunity unavailable to post-vision pruning: it can avoid preprocessing and vision encoding. On A100, this opportunity outweighs a nearly eightfold larger downstream token reduction by the post-vision policy. Reported quality is conditional on examples answered correctly by Full and is not benchmark accuracy.