这篇论文用Facet-Probe测试了18个主流多模态大模型，发现它们对输入顺序都很敏感，最好的模型也错13.4%，提醒我们模型可靠性还不是想象中那么好。
论文提出Facet-Probe审计框架，从选项、证据块、文档排序、图像集、混合模态五个维度测试18个前沿和开源MLLM的排序敏感性。采用贝叶斯项目反应模型分离排序噪声与各维度偏差，发现所有模型均非排序不变，各维度平均翻转率在24%至50%之间。Gemini在温度0下的同序控制显示，验证单元中存在远超解码器噪声的排序超额。最优模型仍有13.4%的试次输出翻转，提示词级缓解措施无法泛化到视觉推理。
Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models
Standard benchmarks for multimodal large language models (MLLMs) score each item on one canonical ordering and miss whether order-irrelevant shuffling changes the answer, a baseline reliability property called for by emerging AI evaluation guidelines. We introduce Facet-Probe, a five-facet audit (option, evidence-chunk, document-rank, image-set, and mixed-modality ordering) of 18 frontier and open-weight MLLMs. A Bayesian item-response model separates ordering noise from per-facet bias, and a same-ordering control estimates the decoder-stochastic floor for observed flips. We find that none of the 18 MLLMs we audit are order-invariant: screened per-facet panel-mean flip rates span 24-50%. A Gemini same-ordering control at temperature 0 estimates a substantial ordering excess over a same-input decoder-noise floor in verified cells. Capability predicts but does not eliminate flips; the best model still flips on 13.4% of trials. In our Gemini mitigation tests, training-free prompt changes are modality-conditional and do not transfer from text to visual reasoning. These results suggest that prompt-level mitigation alone is unlikely to provide general order robustness, motivating future work on training-time and architectural approaches. We propose cross-ordering flip rate as a standard reporting axis for MLLMs.