# Reasoning Multimodal Large Language Model: Data Contamination and Dynamic Evaluation

arXiv: https://arxiv.org/abs/2506.07202 · HTML: https://mlsanity.com/paper/2506.07202/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 21 of its 42 references that are in the map.
- First posted on arXiv 8 Jun 2025.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [MLLM training and visual instruction tuning](https://mlsanity.com/mllm/mllm-training-and-visual-instruction-tuning/)

## Builds on (most cited first, 21 of 21)

- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [Microsoft COCO Captions: Data Collection and Evaluation Server](https://mlsanity.com/paper/1504.00325/) (2015)
- [On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima](https://mlsanity.com/paper/1609.04836/) (2016)
- [MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities](https://mlsanity.com/paper/2308.02490/) (2023)
- [How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suites](https://mlsanity.com/paper/2404.16821/) (2024)
- [SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehension](https://mlsanity.com/paper/2307.16125/) (2023)
- [Are We on the Right Way for Evaluating Large Vision-Language Models?](https://mlsanity.com/paper/2403.20330/) (2024)
- [A Survey on Multimodal Large Language Models](https://mlsanity.com/paper/2306.13549/) (2023)
- [Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpus](https://mlsanity.com/paper/2104.08758/) (2021)
- [Video-R1: Reinforcing Video Reasoning in MLLMs](https://mlsanity.com/paper/2503.21776/) (2025)
- [VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning](https://mlsanity.com/paper/2504.08837/) (2025)
- [MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning](https://mlsanity.com/paper/2503.07365/) (2025)
- [VideoChat-R1: Enhancing Spatio-Temporal Perception via Reinforcement Fine-Tuning](https://mlsanity.com/paper/2504.06958/) (2025)
- [InternVideo2.5: Empowering Video MLLMs with Long and Rich Context Modeling](https://mlsanity.com/paper/2501.12386/) (2025)
- [SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models](https://mlsanity.com/paper/2402.05935/) (2024)
- [DyVal: Dynamic Evaluation of Large Language Models for Reasoning Tasks](https://mlsanity.com/paper/2309.17167/) (2023)
- [NPHardEval: Dynamic Benchmark on Reasoning Ability of Large Language Models via Complexity Classes](https://mlsanity.com/paper/2312.14890/) (2023)

## Built on by (most cited first, 2 of 2)

- [KBE-DME: Dynamic Multimodal Evaluation via Knowledge Enhanced Benchmark Evolution](https://mlsanity.com/paper/2510.21182/) (2025)
- [DMC-CF: Dynamic Multimodal CounterFactual QA benchmark for Causal Reasoning](https://mlsanity.com/paper/2605.29339/) (2026)

## Abstract (arXiv)

Multimodal Large Language Models (MLLMs) show impressive vision-language benchmark performance, yet growing concerns about data contamination (test set exposure during training) risk masking true generalization. This concern extends to reasoning MLLMs, often fine-tuned via reinforcement learning from potentially contaminated base models. We propose a novel dynamic evaluation framework to rigorously assess MLLM generalization, moving beyond static benchmarks. Instead of perturbing inputs, we perturb the task itself. Using the same visual input, models are evaluated across a family of tasks (e.g., QA, captioning, question posing, verification) to probe diverse capabilities. This task perturbation reveals whether model performance is robust or reliant on superficial task-specific cues. Our approach is analogous to loss landscape sharpness: models overfit or contaminated for a single task (sharp minima) falter under task shifts, unlike models with generalizable solutions (flatter minima). We developed an automated pipeline with a calibrated judge scoring open-ended generations (captions, questions) using paraphrase and corruption sampling. Applying this framework to leading image/video MLLMs on benchmarks including MME, RealWorldQA, and CVRR-ES, we analyze each model's cross-task "ability vector." We demonstrate that fine-tuning on simulated test data (extreme contamination) drastically sharpens task-specific performance but harms overall generalization. Our dynamic task perturbation offers deeper insights into MLLM generalization, distinguishing genuine understanding from spurious leakage or overfitting.