benchmarks now show that smaller ai models can perform complex multimodal tasks like video understanding and document analysis, making these capabilities more accessible and affordable.