Must be a state of the art model (probably in the top 8), used by millions of people, not just a research project.

The guiding principle is that the market will resolve to YES if humans are able to follow the thinking process the AI is using. This would likely include cases where:

• The AI's thinking is a mixture of modalities (e.g., image, video, and text), as long as they're comprehensible to humans.
  

  This would not include cases where the AI mostly thinks in "neuralese" latent space vectors, recurrent activations, incomprehensible image/video streams, or the model thinks in text but the text switches languages, uses whitespace meaningfully &c. If in doubt I'll make the call.

• Update 2025-10-22 (PST) (AI summary of creator comment): Current SOTA AI models with visible/user-inspectable Chain-of-Thought (CoT) already satisfy the criterion for what counts as a "human-understandable thinking medium."

• Update 2025-10-22 (PST) (AI summary of creator comment): The market will resolve based on whether the criterion is met at the end of 2028 (not "by" the end of 2028). This is a change from the original title wording.