Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification

Benchmark Model Rank Results
optical-character-recognition-ocr-on-ocrbenchUniAR#41Score: 833.0
video-question-answering-on-mvbenchUniAR#57Avg.: 62.3
text-to-image-generation-on-genevalUniAR (original prompts)#64Overall: 0.85
chart-question-answering-on-chartqaUniAR#801:1 Accuracy: 75.9
multimodal-reasoning-on-mmmuUniAR#114Accuracy: 44.3