AI ResearchJul 10, 2026, 4:38 AM

Meet LingBot-World-Infinity: An Open Causal World Model With An Agentic Harness

30-second summary

Ant Group’s Robbyant unit unveiled LingBot-World-Infinity 2.0, a 14B causal video generation model designed as an interactive world simulator with improved long-horizon stability.

TickrWire
Meet LingBot-World-Infinity: An Open Causal World Model With An Agentic Harness
Key takeaways
  • LingBot-World-Infinity 2.0 is a 14B causal video generation model designed as an interactive world simulator.
  • It introduces a Mixture of Bidirectional and Autoregressive (MoBA) attention mask and distribution matching distillation to reduce long-horizon drift.
  • The model includes a Director-Pilot agentic harness for event proposal and execution, improving simulation coherence.
  • The release is open-source, enabling broader research and integration opportunities.
Full story

Robbyant, Ant Group’s embodied-intelligence division, has launched LingBot-World-Infinity 2.0, an open causal world model that functions as an interactive video simulator. The model leverages a 14 billion parameter architecture and introduces a Mixture of Bidirectional and Autoregressive (MoBA) attention mask, combined with distribution matching distillation over extended self-rollout trajectories. This combination specifically targets long-horizon drift, a persistent issue in interactive world models where textures blur and geometries warp over time.

The system is wrapped in a Director-Pilot agentic harness, where a Vision-Language Model (VLM) proposes events and a secondary agent executes actions within the simulated environment. This separation aims to improve coherence and controllability in long-form simulations, addressing a key limitation in prior world models.

The release is positioned as an open model, likely enabling broader experimentation and integration across research and applied domains. Its focus on stability and agentic control suggests potential applications in robotics, autonomous systems, and synthetic data generation for training.

Why this matters
Developers

Provides a new open tool for building stable, long-horizon interactive simulations with agentic control.

Businesses

Potential to enhance synthetic data generation, robotics training, and autonomous system testing.

Investors

Signals progress in embodied AI and world modeling, a key area for future automation and simulation markets.

Everyone

Advances the realism and controllability of AI-generated virtual environments.

Glossary
Causal world model
A model that simulates how the world evolves over time based on actions and events, rather than just generating static outputs.
Long-horizon drift
The degradation of visual and geometric consistency in simulations over extended time periods.
Distribution matching distillation
A training technique that aligns the output distribution of a student model with that of a teacher model over long trajectories.
Sources · 1
Read next
More stories
TickrWireAI News Intelligence

We aggregate, verify, summarise and explain the latest artificial intelligence news from open, legal sources.

Daily AI digest

Top AI stories, summarised, in your inbox each morning.

© 2026 TickrWire. Summaries and analysis are AI-generated and may contain errors.