OpenVLA: An Open-Source Vision-Language-Action Model
Summary
OpenVLA is a 7B-parameter open-source vision-language-action model trained on 970k real-world robot demonstrations, built on a Llama 2 language model and a visual encoder that fuses DINOv2 and SigLIP features. The authors report that it outperforms RT-2-X (55B) by 16.5% in absolute task success rate across 29 tasks, with 7x fewer parameters, and that fine-tuning can outperform from-scratch methods such as Diffusion Policy by 20.4%. [S1]
Claims
- id: C1 claim: 'OpenVLA is a 7B-parameter open-source VLA trained on 970k real-world robot demonstrations.' source: S1 evidence: "a 7B-parameter open-source VLA trained on a diverse collection of 970k real-world robot demonstrations"
- id: C2 claim: 'The authors report it outperforms RT-2-X (55B) by 16.5% absolute success across 29 tasks, with 7x fewer parameters.' source: S1 evidence: "outperforming closed models such as RT-2-X (55B) by 16.5%"
- id: C3 claim: 'They report a 20.4% gain over from-scratch methods such as Diffusion Policy when fine-tuning OpenVLA.' source: S1 evidence: "outperform expressive from-scratch imitation learning methods such as Diffusion Policy by 20.4%"
Verification
- Code available: Comments and HTML point to https://openvla.github.io/, which returned HTTP 200 on 2026-11-12. Not run.
- Reproduced: not reproduced (2026-11-12). Code not run.
- Related work: Live notes 2310-08864-open-x-embodiment and 2307-09288-llama-2.
- Limits: The 16.5% and 20.4% figures are author-reported (v3, 5 Sep 2024). This note cites the project site named in the paper.
- License: CC BY 4.0 on the abs page (2026-11-12; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 13 Jun 2024 (v1); last revised 5 Sep 2024 (v3). Sources: [S1] https://arxiv.org/abs/2406.09246, v3 (5 Sep 2024). checked_on: 2026-11-12.
When to apply
Use when asking how an open vision-language-action model was trained on large multi-robot demonstration data.
Facts
- id: F1 claim: 'OpenVLA is a 7B open-source vision-language-action model trained on 970k real-world robot demonstrations.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F2 claim: 'It combines a Llama 2 language model with a visual encoder that fuses DINOv2 and SigLIP features.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F3 claim: 'The authors report a 16.5% absolute success-rate gain over RT-2-X (55B) across 29 tasks, with 7x fewer parameters.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
Sources
- [S1] arXiv 2406.09246, https://arxiv.org/abs/2406.09246. License: CC BY 4.0. v3 (5 Sep 2024). Accessed 2026-11-12.
License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.
License: CC-BY-4.0 · Summary by iinventedlight (AI-assisted). Source: Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, and 15 coauthors (listed on the arXiv abs page), OpenVLA: An Open-Source Vision-Language-Action Model (CC BY 4.0).
Cite this
'iinventedlight (AI-assisted)'. 2026. "OpenVLA: An Open-Source Vision-Language-Action Model." iinventedlight Library, iil:note:kt3zcfcc, version of 2026-11-12. https://iinventedlight.com/library/notes/2406-09246-openvla/. License: CC-BY-4.0.
@misc{iil_note_kt3zcfcc,
author = {{'iinventedlight (AI-assisted)'}},
title = {{OpenVLA: An Open-Source Vision-Language-Action Model}},
howpublished = {iinventedlight Library},
year = {2026},
date = {2026-11-12},
url = {https://iinventedlight.com/library/notes/2406-09246-openvla/},
note = {Permanent ID iil:note:kt3zcfcc. License: CC-BY-4.0.},
}
Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.
For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2406-09246-openvla.md.