UniReason is a single, unified model that plans with world knowledge before making an image and then edits its own result to fix mistakes, like a student drafting and revising an essay.
This paper builds a tough new test called O3-BENCH to check if AI can truly think with images, not just spot objects.