Research Preview · 2026

Building the
Agentic World
Stack

XGEN Labs is pioneering the infrastructure for agents that can perceive, act, remember, and evolve inside continuous, interactive worlds.

Explore Modules See the Stack
Benchmark Layer
UniWorldBench — real spatio-temporal evaluation
Model Layer
World Context Model · World Agency Model
Harness Layer
Agent Harness · Engine Harness
Unified Model Layer
Converged understanding, generation & action
Long-Term Direction

From Fragmented
Models to a Unified
World Model

The next generation of AI won't be a collection of separate VLMs, video generators, audio models, and agent policies. They will converge into a single, shared semantic world space — where understanding, generation, intent formation, and interactive action happen inside one unified representation.

That is what XGEN Labs is building toward.

Unified
World
Model
VLM Video Audio Policy