Simulating the
World's Intelligence

Steerable world models for reinforcement learning

Simulating environment dynamics from grounded agent trajectories

Studying transfer to unseen environments

Training data drawn from expert Figma workflows

Preserving design intent when converting workflows into trajectories

Improving agent performance across computer-use environments

Testing long-term memory in dynamic work environments

Tracking changing information across connected platforms

Revealing failures in retaining and applying prior context

Testing multi-turn search with incomplete information

Resolving ambiguous requests through agent interaction

Spanning text, image, audio, and video

Investigating failures across complex agent workflows

Human-annotated traces from software and research tasks

Testing models’ ability to locate and explain errors

Detecting reward hacking in coding environments

Comparing intended behavior with exploitative agent trajectories

Testing whether models recognize shortcuts that undermine task goals

Testing strategy discovery through repeated gameplay

Challenging agents to improve beyond task completion

Measuring planning and execution under time pressure

Testing financial reasoning against real company documents

Expert-curated questions with answers and supporting evidence

Revealing limitations in financial question answering