Research
Ψ-0.6 unified vision-language world model reasons over words, depth, motion and imagined scenes
Ψ-0.6 is presented as one model acting as perceiver, VLM, world model, reasoner and policy, with RL learning which representation to reason through. The authors report it beats leading open-weight models on MindCube zero-shot and outperforms other world models on FlowBench, and that training on sim…
Every outlet that ran it 1 article
r/MachineLearningCommunityLead
Ψ-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model [R]