Is Liber-0 state of the art?Pending benchmarks.Announced Nov 3, 2026; no independent results yet.
LiberAI's World Action Model for zero-shot robot manipulation, pretrained on over 2.2M hours of human experience. It jointly predicts future observations and actions, using a Mixture-of-Transformers with language, vision and action experts. Data comes from LiberAI's LiberSense glove and it runs on the LEO robot. No weights or API announced.
The details so far
- Lab
- LiberAI
- Announced
- Nov 3, 2026
- API
- Not yet · None announced
- Open weights
- Unknown · Not stated
- Type
- World Action Model for robot manipulation; not an LLM
- Architecture
- Mixture-of-Transformers with language, vision and action experts; jointly predicts future observations and actions
- Pretraining
- Over 2.2M hours of human experience
- Hardware
- LiberSense Glove (0.3 mm thick, 131 g) for data capture; LEO robot
- Planner
- Can optionally pair with GPT-6 Astra for task planning
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: 36.0% overall zero-shot success over 600 trials on 60 tasks proposed by GPT-6 (pick-and-place 63.5%, other short tasks 34.0%, multi-step 10.5%), without task-specific robot demonstrations.
Lab claim: 83.6 on pick-and-place and 73.8 on other manipulation with the full training recipe.
Sources
LABLiberAI announcement on Xx.com/liberai_robot/status/2117803284039905456 · Nov 3, 2026LABLiber-0 technical blogliberai.com/blog/liber0Tracking ends Feb 2, 2027 unless the API or weights become available.