โ† OSWorld-Science

Trajectory Showcase

Ten tasks in five applications (QuPath, ANSYS Fluent, SAOImage DS9, Praat and QGIS), each replayed step by step for one fully credited run from the main results, by seven models in all. Every frame is the agent's own screenshot; the text is the model's own narration at that step (its reply, or the comments in its code). Use โ† โ†’ to move through the run.