Sign in
Pending benchmarksTracking since Nov 4, 2026

Is Verne V0.5 state of the art?Pending benchmarks.Announced Nov 4, 2026; no independent results yet.

Verne Robotics' open-weight world action model for industrial dual-arm mobile robots: a video transformer and an action transformer trained jointly, emitting 30 Hz action chunks.

Announced facts

The details so far

Lab
Verne Robotics
Announced
Nov 4, 2026
API
Not yet
Open weights
Published
Parameters
8.22B (8,216,402,458)
Type
World action model (policy layer of a four-layer robot stack)
Checkpoints
v0.5-base, v0.5-libero and v0.5-robotwin
Weights access
Gated on Hugging Face (manual approval), custom license
Output
30 Hz action chunks with end-effector poses, gripper commands, base position and progress
Fine-tuning
Hosted fine-tuning platform launched alongside the weights

Tags marked unsure are not stated on the lab's own pages.

Next steps

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)
Self-reported

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: 99.3% average success on LIBERO (99.2% on LIBERO-Long) after fine-tuning on its training demonstrations.

Lab claim: 81.0% success in clean and 73.4% in randomized RoboTwin 2.0 scenes.

Lab claim: highest LIBERO and RoboTwin 2.0 results among open-weight models from US companies as of Nov 1, 2026.

Receipts

Sources

LABIntroducing V0.5vernerobotics.com/research/verne-v0-5 · Nov 4, 2026LABVerne Robotics on X: V0.5 open-weight world action modelx.com/VerneRobotics/status/2118029255804129448 · Nov 4, 2026REPOvernerobotics/v0.5-base on Hugging Facehuggingface.co/vernerobotics/v0.5-base

Tracking continues until an independent board lists it.