PhAI Labs

PhAI Labs, with The Chinese University of Hong Kong, University of Oxford, Stanford University, and Princeton University, has released JEPA-Anything, a cross-domain world model framework. Its orthogonal predictive decomposition (OPF) splits the target state into complementary subspaces, with separate prediction branches handling each. The team evaluated the model on visual, biological, clinical, control, molecular, physical-field, and weather systems. They report it outperforms standard JEPA on multiple prediction tasks and that internal factors yield scientific clues for cancer-cell intervention and analysis of planetary orbit regularities.

ScienceIDE, the first open-source release from AItonomy Foundation, packages 27 scientific codebases into 64 executable environments with 2,812 tasks and 1,076 scientific checks. Sponsored by PhAI Labs and Alibaba Qwen, the project has domain experts define examples and acceptance criteria, then works with AI to turn verified interactions into data for evaluation, supervised fine-tuning and reinforcement learning. It also proposes seven task categories.

The team released ScienceIDE-Hard, an 85-task public benchmark drawn from PLUTO, Athena++, MITgcm, LAPS and PHANTOM. Across 15 models, the strongest passed 67% of tasks under the scientific-equivalence criterion. Licence terms were not stated.