SurgKnow-Bench
The boundary of generalist models in surgery.
Surgical Intelligence
Agents that watch an operation, call tools to solve what they cannot alone, and learn from their own attempts.
We measure what general multimodal models already know about surgery, on their own.
We give the agent tools to perceive the video and retrieve knowledge, and measure how much of the gap it closes.
The agent learns from its own trajectories in the Gym.
The boundary of generalist models in surgery.
Where agents watch surgical video, call tools and leave a full trajectory.
Curated surgical knowledge, built from what a surgical resident studies.
The reference agent. Version 1 reproduces SurgKnow-Bench.
Agents that learn from their own trajectories.