Ψ-0.6 Introduces Adaptive Streams of Thought for Unified Vision-Language World Modeling

·2026.10.28 01:09

Key point

The model surpasses leading open-weight models on MindCube spatial reasoning and outperforms other world models on FlowBench motion prediction.

Ψ-0.6 is a unified model that reasons in streams of thought, including words, depth, motion, camera pose, and imagined scenes.
The model surpasses leading open-weight models on complex spatial reasoning in the MindCube benchmark (zero-shot).
On FlowBench, Ψ-0.6 outperforms other world models at predicting how objects move.
Trained on simulated rooms, the model transfers to real robot tabletops.
Reinforcement learning (RL) is used to determine which representation to reason through.
The model functions as a perceiver, VLM, world model, reasoner, and policy simultaneously.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.