NEO-VLA: Encoder-Free Vision-Language-Action Model Released

The NEO team has released NEO-VLA, an encoder-free native vision-language-action model that maps pixels directly to actions, challenging the assumption that VLAs need semantically distilled visual representations. Paper, code and project page are all open-sourced.

2026-11-11 ~ 2026-11-11 · 2 related posts

• NEO-VLA: encoder-free native VLA turns pixels into actions, beats encoder-based baselines — liuziwei7 · 2026-11-11
• NEO-VLA paper: native visual beats encoder features for robot action, data cleaned from 2.2M episodes — liuziwei7 · 2026-11-11