Catalyzing the
Advancing AI and physical science for autonomous research.
Our Portfolio
Open-source ecosystems for autonomous discovery
LabOS
AI-XR co-scientist platform uniting multimodal perception, agents, and robotics to observe, reason, and collaborate with humans in real wet-labs.
CRISPR-GPT
AI agent system that automates complex gene-editing experiment design, robust protocol generation, and execution through natural language.
LabClaw
A comprehensive skill library and tool-creation module designed to power autonomous research and seamless physical execution in laboratories.
RNAGenesis
Generalist foundation model for RNA therapeutics that unifies deep sequence understanding, de novo RNA design, and 3D structure prediction.
MedOS
An advanced AI-XR-Cobot platform built specifically for complex clinical reasoning, intelligent diagnostics, and interactive medical workflows.
Physics Supernova
Elite AI agent system that scored 23.5/30 on theoretical questions at the 2025 International Physics Olympiad, surpassing human gold medalists.
LabOS
AI-XR co-scientist platform uniting multimodal perception, agents, and robotics to observe, reason, and collaborate with humans in real wet-labs.
CRISPR-GPT
AI agent system that automates complex gene-editing experiment design, robust protocol generation, and execution through natural language.
LabClaw
A comprehensive skill library and tool-creation module designed to power autonomous research and seamless physical execution in laboratories.
RNAGenesis
Generalist foundation model for RNA therapeutics that unifies deep sequence understanding, de novo RNA design, and 3D structure prediction.
MedOS
An advanced AI-XR-Cobot platform built specifically for complex clinical reasoning, intelligent diagnostics, and interactive medical workflows.
Physics Supernova
Elite AI agent system that scored 23.5/30 on theoretical questions at the 2025 International Physics Olympiad, surpassing human gold medalists.
Alita: Self-Evolving Agents
Minimalist self-evolving generalist AI agent dynamically generating capabilities via Model Context Protocols to achieve 75.42% on the GAIA benchmark.
TraceRL
Trajectory-aware reinforcement learning framework enabling diffusion-based language models to outperform larger autoregressive models in math.
MATH-Perturb
Rigorous benchmark testing whether LLMs truly understand mathematical reasoning or rely on memorization by applying hard problem perturbations.
Alita: Self-Evolving Agents
Minimalist self-evolving generalist AI agent dynamically generating capabilities via Model Context Protocols to achieve 75.42% on the GAIA benchmark.
TraceRL
Trajectory-aware reinforcement learning framework enabling diffusion-based language models to outperform larger autoregressive models in math.
MATH-Perturb
Rigorous benchmark testing whether LLMs truly understand mathematical reasoning or rely on memorization by applying hard problem perturbations.
Alita: Self-Evolving Agents
Minimalist self-evolving generalist AI agent dynamically generating capabilities via Model Context Protocols to achieve 75.42% on the GAIA benchmark.
TraceRL
Trajectory-aware reinforcement learning framework enabling diffusion-based language models to outperform larger autoregressive models in math.
MATH-Perturb
Rigorous benchmark testing whether LLMs truly understand mathematical reasoning or rely on memorization by applying hard problem perturbations.
Alita: Self-Evolving Agents
Minimalist self-evolving generalist AI agent dynamically generating capabilities via Model Context Protocols to achieve 75.42% on the GAIA benchmark.
TraceRL
Trajectory-aware reinforcement learning framework enabling diffusion-based language models to outperform larger autoregressive models in math.
MATH-Perturb
Rigorous benchmark testing whether LLMs truly understand mathematical reasoning or rely on memorization by applying hard problem perturbations.