Griffin LabsSingapore2026
Griffin Alpha-S
An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.
Robotsnot stated
griffinlabs.ai
Key figures as stated by the developer
5 of 6 stated · 1 gap-
Model type A
vision-language-action -
Controls B
arms and hands -
Inputs A
cameralanguagerobot state - Access B open download
- Parameters not stated
- Base model C Qwen3-VL-4B-Instruct
How to read the figures
- 50HzA
- The letter in the grey square after a figure points to the source the figure came from. Pages with several sources carry several letters.
Versions and notes in the key figures (6)
Model type
Developer wording: "Griffin Alpha-S is a vision-language-action (VLA) model by" Griffin Labs.
Controls
Blog post: "One set of weights supports multiple robots and both joint-space and end-effector actions." "One set of released Griffin Alpha-S weights running CleanBench on bimanual Piper arms."
Inputs
Repository README: "Both predict a 50-step action chunk from up to three camera images, the proprioceptive state, and a language instruction". Proprioceptive state = robot state.
Access
Blog post: "Open-weights release, September 2026"; the weights are downloaded with the lerobot tools.
Parameters
Note Griffin Labs gives one number for one part: "a 910M-parameter flow-matching action expert attached layer-by-layer (mixture-of-transformers) to a Qwen3-VL-4B backbone".
Base model
Model card header: "base_model: Qwen/Qwen3-VL-4B-Instruct".
Full specifications
9 of 13 fields stated| Field | Value | Source and caveats |
|---|---|---|
| Input and output | ||
| Model type | vision-language-action | A Developer wording: "Griffin Alpha-S is a vision-language-action (VLA) model by" Griffin Labs. |
| Inputs | camera language robot state | A Repository README: "Both predict a 50-step action chunk from up to three camera images, the proprioceptive state, and a language instruction". Proprioceptive state = robot state. |
| Controls | arms and hands | B Blog post: "One set of weights supports multiple robots and both joint-space and end-effector actions." "One set of released Griffin Alpha-S weights running CleanBench on bimanual Piper arms." |
| Degrees of freedom controlled | not stated | Note Model card: the base uses a "canonical 32-dimensional action vector", with each embodiment's actions in the leading slots; this is the action width, not a count of controlled joints. C |
| Size and speed | ||
| Parameters | not stated | Note Griffin Labs gives one number for one part: "a 910M-parameter flow-matching action expert attached layer-by-layer (mixture-of-transformers) to a Qwen3-VL-4B backbone". C |
| Where it runs and what it is built on | ||
| Base model | Qwen3-VL-4B-Instruct | C Model card header: "base_model: Qwen/Qwen3-VL-4B-Instruct". |
| Openness | ||
| Access | open download | B Blog post: "Open-weights release, September 2026"; the weights are downloaded with the lerobot tools. |
| License | CC BY-NC-SA 4.0 | C Model card header for the weights: "license: cc-by-nc-sa-4.0". "The plugin code is Apache-2.0." |
| Free or paid | open, non-commercial only | C Weights licence, model card header: "license: cc-by-nc-sa-4.0". |
| The developer’s own claims | ||
| Learns new tasks | needs retraining | C Developer claim: the pre-trained base is "meant as the starting point for fine-tuning on your own robot". |
| Works across robots | yes | B Developer claim: "One set of weights supports multiple robots". 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. |
Not stated: Control rate, On robot or cloud
Notes
- Griffin Labs calls Griffin Alpha a robot foundation model and releases its open core, Griffin Alpha-S.
- Griffin Labs published the open weights in September 2026.
- The Griffin Labs site is signed Griffin Labs Pte. Ltd. and lists its open roles in Singapore.
Sources
-
A
github.
com/ · Retrievedgriffinlabs-ai/ alpha-s -
B
Primary source. griffinlabs.
ai/ · Retrievedblog/ griffin-alpha-s -
C
huggingface.
co/ · Retrievedgriffinlabs/ Griffin-Alpha-S
One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.