MapTab: Are MLLMs Ready for Multi-Criteria Route Planning in Heterogeneous Graphs?

MapTab is a specialized research benchmark targeting a specific gap in current MLLM evaluation: the intersection of visual perception and complex constraint-based graph reasoning. While the project is very young (8 days old) and currently lacks public stars, the 8 forks suggest initial interest from the academic community following its arXiv publication. Its defensibility is low because it is a evaluation framework rather than a product; its value lies entirely in its adoption rate by other researchers (citation moat). It faces 'medium' frontier risk because labs like Google (via Gemini/Maps) and OpenAI are aggressively pursuing 'World Models' and spatial intelligence, which would naturally subsume the capabilities this benchmark measures. Its primary competitors are general multimodal benchmarks like MMMU or ChartQA, but MapTab carves a niche in multi-criteria decision-making. The moat is essentially the difficulty of curating high-quality, heterogeneous graph-based visual tasks, but this is vulnerable to automated synthetic data generation pipelines developed by larger labs.