mlsanityData to 14 Oct 2026

Token-Efficient Multimodal Reasoning via Image Prompt Packaging

arXiv 2604.02492 ↗First posted 2 Apr 2026cs.CV cs.AIcited by 0

On the map: LLM reasoning → Multimodal reasoning

also in 1 more
Abstract (arXiv)

Deploying large multimodal language models at scale is constrained by token-based inference costs, yet the cost-performance behavior of visual prompting strategies remains poorly characterized. We introduce Image Prompt Packaging (IPPg), a prompting paradigm that embeds structured text directly into images to reduce text token overhead, and benchmark it across five datasets, three frontier models (GPT-4.1, GPT-4o, Claude 3.5 Sonnet), and two task families (VQA and code generation). We derive a cost formulation decomposing savings by token type and show IPPg achieves 35.8--91.0% inference cost reductions. Despite token compression of up to 96%, accuracy remains competitive in many settings, though outcomes are highly model- and task-dependent: GPT-4.1 achieves simultaneous accuracy and cost gains on CoSQL, while Claude 3.5 incurs cost increases on several VQA benchmarks. Systematic error analysis yields a failure-mode taxonomy: spatial reasoning, non-English inputs, and character-sensitive operations are most vulnerable, while schema-structured tasks benefit most. A 125-configuration rendering ablation reveals accuracy shifts of 10--30 percentage points, establishing visual encoding choices as a first-class variable in multimodal system design.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 13 of its 20 references that are in the map. First posted on arXiv 2 Apr 2026.

Builds on

13 of 20 references.

References · 13

  1. 2023Gemini: A Family of Highly Capable Multimodal Modelscited by 872
  2. 2023Do All Languages Cost the Same? Tokenization in the Era of Commercial Language Modelscited by 192
  3. 2025DeepSeek-OCR: Contexts Optical Compressioncited by 183
  4. 2025MLLMs are Deeply Affected by Modality Biascited by 40
  5. 2022Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimizationcited by 32
  6. 2025A Closer Look at Multimodal Representation Collapsecited by 28
  7. 2024Unveiling Typographic Deceptions: Insights of the Typographic Vulnerability in Large Vision-Language Modelcited by 22
  8. 2025When Language Overrules: Revealing Text Dominance in Multimodal Large Language Modelscited by 18
  9. 2025Texture or Semantics? Vision-Language Models Get Lost in Font Recognitioncited by 11
  10. 2025Text or Pixels? It Takes Half: On the Token Efficiency of Visual Text Inputs in Multimodal LLMscited by 8
  11. 2025Some Modalities are More Equal Than Others: Decoding and Architecting Multimodal Integration in MLLMscited by 7
  12. 2024MMLF: Multi-modal Multi-class Late Fusion for Object Detection with Uncertainty Estimationcited by 3
1 more
  1. 2025Read or Ignore? A Unified Benchmark for Typographic-Attack Robustness and Text Recognition in Vision-Language Modelscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026When Prompts Become Pixels: Prompt-Region Grounding for Multimodal Reasoning
  2. 2026TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs
  3. 2024POEM: Interactive Prompt Optimization for Enhancing Multimodal Reasoning of Large Language Models
  4. 2026Visual Tokens on Demand: Visual Budgeting for Scalable and Efficient Multi-Image Reasoning in Multimodal LLMs
  5. 2024PAR: Prompt-Aware Token Reduction Method for Efficient Large Multimodal Models
  6. 2024Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings
  7. 2024Plug-and-Play Grounding of Reasoning in Multimodal Large Language Models
  8. 2026Test-Time Computing for Referring Multimodal Large Language Models