From Reasoning to Action: Benchmarking Frontier AI Models on LIBERO-10 Robot Manipulation Tasks
GPT-6 Astra led the robot manipulation benchmark at 44% success, narrowly ahead of Gemini-4 Argon’s 43%.
The researchers tested six frontier models across three reasoning effort levels in 1,800 LIBERO-10 episodes. Models controlled the robot through visual observations, robot state data and structured commands, without task-specific fine-tuning or pretrained manipulation skills. More reasoning generally improved performance, with gains varying by model and adding inference cost. Object placement proved easier than precise spatial arrangements and complex multi-step interactions.
ArXiv · AI/CL/LG's note
The researchers tested six frontier models across three reasoning effort levels in 1,800 LIBERO-10 episodes. Models controlled the robot through visual observations, robot state data and structured commands, without task-specific fine-tuning or pretrained manipulation skills. More reasoning generally improved performance, with gains varying by model and adding inference cost. Object placement proved easier than precise spatial arrangements and complex multi-step interactions.
ArXiv · AI/CL/LG's note
score 5