Open weightsStanford / UC Berkeley / Toyota Research Institute/10 trustVerified

OpenVLA

Name: OpenVLA
Rating: 8

The open vision-language-action model for generalist robot control.

About

OpenVLA is a 7B-parameter open-source vision-language-action (VLA) model: it takes a camera image plus a language instruction and outputs robot actions. It combines a Llama 2 backbone with a visual encoder fusing DINOv2 and SigLIP features, trained on 970K real robot demonstrations from the Open X-Embodiment dataset.

It beat the closed 55B RT-2-X by 16.5% absolute success rate across 29 manipulation tasks with 7x fewer parameters, and it supports parameter-efficient fine-tuning to new robots. Checkpoints, training code, and data are fully open on Hugging Face — it became the default starting point for open robotics research.

Specs

Context window

Not applicable in the LLM sense — takes one camera image plus a text instruction per step and outputs a 7-DoF action.

Pricing

Free — open weights and code.

Free quota

Full checkpoints on Hugging Face; runs on a single consumer GPU with 4-bit quantization.

API access

Hugging Face checkpoints; PyTorch training and inference pipeline

Weights

Open weights

Released

June 1, 2024

Benchmarks

Outperformed closed RT-2-X (55B) by 16.5% absolute task success rate across 29 tasks and multiple robot embodiments (paper: arXiv 2406.09246).

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

automationresearch

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: default starting point for open robotics VLA research; paper-era maintenance; community fine-tunes actively (e.g. OpenVLA-OFT); fully open: data, weights, and training code. Manual review: strong published results vs RT-2-X; real-robot performance depends heavily on your embodiment and data. License: Open weights; confirm license on the Hugging Face model card..

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 8.0/10
Maintenance · recency, releases, issue load: 7.0/10
Openness · license + docs: 10.0/10
Manual review · sanity read + red-flag check: 8.0/10

· default starting point for open robotics VLA research
· paper-era maintenance; community fine-tunes actively (e.g. OpenVLA-OFT)
· fully open: data, weights, and training code
· manual review: strong published results vs RT-2-X; real-robot performance depends heavily on your embodiment and data