I’m seeing really good reviews of OpenAI’s new image model, but I find that even with Thinking, it hallucinates quite a few data points, which, therefore, makes the result unusable no matter how readable and well-formatted it is… Am I missing something?

→ View original post on X — @dotcsv,