Verne Robotics · robot AI model
V0.5
United States · 2026
Sources retrieved: 9 November 2026 (A, B, C)
- Model type
- World action modelSource and note: Model typeB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Developer wording: “V0.5 is a world action model (WAM): it learns to predict both what the robot will see next and what it should do.”
- What it controls
- Arms and hands Whole bodySource and note: What it controlsA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “it returns the next 1.07 seconds of motion: 32 targets at 30 Hz for both end effectors, both grippers and the mobile base.” Verne's page: “Each action chunk moves the base and both arms together.”
- Inputs
- Camera Language Robot stateSource and note: InputsA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “You give it three camera views, the robot's state and an instruction in plain English.”
- Access
- Open weights (download)Source and note: AccessB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Verne's page: “Build on V0.5. Fine-tune it on your own demonstrations, or download the weights.” The Hugging Face checkpoints are gated; README: “hf auth login # accept the license on the checkpoint page first”
- Parameters
- 8.2BSource and note: ParametersA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “The weights are 8.2B parameters in bf16 (17 GB).”
- Base model
- Wan2.2Source and note: Base modelB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Verne's page: “V0.5 starts from the Wan2.2 video model.”
- Licence document
- Verne V0.5 Model LicenseSource and note: Licence documentA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “V0.5 is released under the Verne V0.5 Model License. For commercial use, please contact partnership@vernerobotics.com.” The licence text: “Any person or organization may make Research Use of the Materials, at any Annual Revenue.” And: “You may make Commercial Use of the Materials only if Your Annual Revenue is at or below the Revenue Threshold.” The Revenue Threshold is US$5,000,000 of annual revenue.
Specifications
10 of 14 stated 4 not stated
1 field hidden 3 not-stated fields with a note stay ·
How to read the figuresShow guideClose
- not statedNo source
- 0Reported as zero
- ASource
Select a source marker for source details and notes.
Release1/1 disclosed
ReleasedNovember 2026
Source and note: ReleasedB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. [A] The page is dated: “Verne Team November 2026”.
Input and output3/4 disclosed
Model typeWorld action model
Source and note: Model typeB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Developer wording: “V0.5 is a world action model (WAM): it learns to predict both what the robot will see next and what it should do.”
InputsCamera Language Robot state
Source and note: InputsA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “You give it three camera views, the robot's state and an instruction in plain English.”
What it controlsArms and hands Whole body
Source and note: What it controlsA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “it returns the next 1.07 seconds of motion: 32 targets at 30 Hz for both end effectors, both grippers and the mobile base.” Verne's page: “Each action chunk moves the base and both arms together.”
Degrees of freedom controlledNot stated
Size and speed2/2 disclosed
Parameters8.2B
Source and note: ParametersA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “The weights are 8.2B parameters in bf16 (17 GB).”
Control rate500Hz
Source and note: Control rateB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. The figure is the separate controller layer: “It runs in real time at 500 Hz, independent of model inference.” The model itself outputs 30 Hz targets: “it returns the next 1.07 seconds of motion: 32 targets at 30 Hz for both end effectors, both grippers and the mobile base.”
Where it runs and what it is built on1/2 disclosed
On robot or cloudNot stated
Source and note: On robot or cloudA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “You need Linux, Python 3.10 or newer, an NVIDIA GPU with at least 24 GB, 32 GB of RAM and 35 GB of disk.”
Base modelWan2.2
Source and note: Base modelB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Verne's page: “V0.5 starts from the Wan2.2 video model.”
Openness2/3 disclosed
AccessOpen weights (download)
Source and note: AccessB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Verne's page: “Build on V0.5. Fine-tune it on your own demonstrations, or download the weights.” The Hugging Face checkpoints are gated; README: “hf auth login # accept the license on the checkpoint page first”
Licence documentVerne V0.5 Model License
Source and note: Licence documentA github.com/vernerobotics/v0.5 · Retrieved 9 November 2026Note. README: “V0.5 is released under the Verne V0.5 Model License. For commercial use, please contact partnership@vernerobotics.com.” The licence text: “Any person or organization may make Research Use of the Materials, at any Annual Revenue.” And: “You may make Commercial Use of the Materials only if Your Annual Revenue is at or below the Revenue Threshold.” The Revenue Threshold is US$5,000,000 of annual revenue.
Licence typeNot stated
Source and note: Licence typeC github.com/vernerobotics/v0.5/blob/main/LICENSE · Retrieved 9 November 2026Note. The licence is free for research and for commercial use at or below US$5,000,000 annual revenue: “You may make Commercial Use of the Materials only if Your Annual Revenue is at or below the Revenue Threshold.”
The developer’s own claims1/2 disclosed
Learns new tasksNeeds retraining
Source and note: Learns new tasksB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Developer claim. Verne's page: “Learn a new task, or a new step of a longer task, from under two hours of demonstrations and human corrections, building on what the model already knows, and recover from errors.” README: “Fine-tune first. The base checkpoint has not seen your robot, cameras or tasks.”
Works across robotsNot stated
Source and note: Works across robotsB vernerobotics.com/research/verne-v0-5 · Retrieved 9 November 2026Note. Verne's page: “Demonstrations from different rigs then share one description of motion, whatever their joints”. README: “Fine-tune first. The base checkpoint has not seen your robot, cameras or tasks.”Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.
Notes
Verne says V0.5 is the policy layer of a four-layer stack: a high-level reasoning layer, the policy layer, an execution layer and a controller.
Verne says V0.5 acts in end-effector space rather than joint space, so demonstrations from different robots share one description of motion.
Verne says it pretrained V0.5 on over 1.7 million demonstrations from data vendors, open-source datasets and its own collection.
Verne describes itself as a US company; its site footer names the company Verne Labs, Inc.
Sources
One line per source, not per figure: a source has one retrieval date.
- Agithub.com/vernerobotics/v0.5Retrieved 9 November 2026
- Bvernerobotics.com/research/verne-v0-5Main source. Retrieved 9 November 2026
- Cgithub.com/vernerobotics/v0.5/blob/main/LICENSERetrieved 9 November 2026
