Robbyant open-sources LingBot-World 2.0: 1.3B world model runs interactive worlds on one consumer GPU
dair_ai · x · 2026-09-10
Robbyant announced the open-sourcing of three new versions of its real-time interactive world model, LingBot-World 2.0:
- LingBot-World 2.0 Small (1.3B): generates an interactive open world in real time on a single consumer GPU, scaling to 720p/60fps in the full setup.
- Bidirectional and Causal Pretrain variants are also open-sourced.
- The generated world isn't a seconds-long clip: it runs for hours, responds to every action (attack, cast spells, shoot, summon storms) and keeps evolving through AI agents.
dair-ai notes this follows the same pattern seen with language and image models: scale capabilities first, then distill down to hardware people actually own.
More from Multimodal
- Open-Source Ref2V Workflow Auto-Transcribes Media and Formats Prompts for H3 Video Generation — bstr3k · 2026-09-10
- fal launches Multi-angle for H3 Max: any camera angle from one image in under 3 seconds — gorkem · 2026-09-10
- Valeo fits scaling laws for video diffusion using 5,500 hours of driving footage — abursuc · 2026-09-10
- Zero-cost fix for Minimax H3 plastic skin: drop contrast before VAE decode — Cequejedisestvrai · 2026-09-10
- 15-second single-pass AI bike chase: three craft lessons from a MiniMax H3 video — LudovicCreator · 2026-09-10
- Seven-episode AI-generated Dragon Ball fan series drops episode 7 teaser — Stunning_Act_1530 · 2026-09-10